Next Article in Journal
Autocorrelation and Parameter Estimation in a Bayesian Change Point Model
Previous Article in Journal
Solitons in Neurosciences by the Laplace–Adomian Decomposition Scheme
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Review

A Review on Nature-Inspired Algorithms for Cancer Disease Prediction and Classification

1
School of Advanced Science and Languages, VIT Bhopal University, Kothrikalan, Sehore 466114, India
2
School of Computing Science and Engineering, VIT Bhopal University, Kothrikalan, Sehore 466114, India
*
Author to whom correspondence should be addressed.
Mathematics 2023, 11(5), 1081; https://doi.org/10.3390/math11051081
Submission received: 3 January 2023 / Revised: 18 February 2023 / Accepted: 20 February 2023 / Published: 21 February 2023

Abstract

:
In the era of healthcare and its related research fields, the dimensionality problem of high-dimensional data is a massive challenge as it is crucial to identify significant genes while conducting research on diseases like cancer. As a result, studying new Machine Learning (ML) techniques for raw gene expression biomedical data is an important field of research. Disease detection, sample classification, and early disease prediction are all important analyses of high-dimensional biomedical data in the field of bioinformatics. Recently, machine-learning techniques have dramatically improved the analysis of high-dimension biomedical data sets. Nonetheless, researchers’ studies on biomedical data faced the challenge of vast dimensions, i.e., the vast features (genes) with a very low sample space. In this paper, two-dimensionality reduction methods, feature selection, and feature extraction are introduced with a systematic comparison of several dimension reduction techniques for the analysis of high-dimensional gene expression biomedical data. We presented a systematic review of some of the most popular nature-inspired algorithms and analyzed them. The paper is mainly focused on the original principles behind each of the algorithms and their applications for cancer classification and prediction from gene expression data. Lastly, the advantages and disadvantages of nature-inspired algorithms for biomedical data are evaluated. This review paper may guide researchers to choose the most effective algorithm for cancer classification and prediction for the satisfactory analysis of high-dimensional biomedical data.

1. Introduction

The ML-based model with high-dimensional biomedical data, for early prediction and categorization of cancer is a new area of bioinformatics research [1,2,3]. High-dimensional gene expression biomedical data produced a large amount of gene expression with a small sample size. Machine learning based soft computing techniques with high-dimensional gene expression data have become a powerful tool for early cancer prediction, which helps to understand the causes of cancer and opens the door to the development of novel treatments. Due to the high-dimensionality of gene expression data, it is difficult to find a small number of useful genes for disease classification and prediction with high accuracy. The chosen informative genes help in improving treatment approaches and early disease diagnosis [4,5,6]. Recently, various researchers proposed different soft computing-based frameworks for finding important genes from high-dimensional data that helped train the model with high accuracy for further analysis. Researchers used different ML techniques for gene selection (dimension reduction algorithms) and found significant features (genes) that are needed for learning purposes. Since each gene expression data set has unique characteristics, defining an ideal decision framework for dimension reduction is a challenging task in the fields of machine learning and medical science. Dimension reduction algorithms, including feature extraction and feature selection, are commonly used to solve this problem. Feature extraction algorithms are statistical methods that transform an existing input surface into a new one by generating new features [7,8,9]. On the other hand, in dimension reduction, feature selection algorithms have been applied to gene expression datasets to select ideal discriminating genes called biomarkers. Nowadays, to address the issue of dimension reduction, many researchers use a variety of hybrid approaches with various combinations to speed up computation and gain the advantages of the various dimension reduction techniques [10]. Researchers used various combinations of algorithms in accordance with the demands of various data sets by integrating various feature selection and extraction techniques. Utilizing the most effective techniques is essential because resources and time are both finite. Nature-inspired metaheuristic algorithms have recently gained popularity as a tool for discovering significant features from large and complex biomedical datasets, as compared to other algorithms. Nature-inspired techniques include a wide range, from local learning techniques to global search solutions. Exploration and exploitation are the two key working processes in every nature-inspired metaheuristic algorithm. The exploitation process is looking for the best option from the currently available options, while exploration is discovering a novel solution on a large scale. Nature-inspired algorithms are founded on the diversification principle and take some clues from the environment to solve problems. Several nature-inspired techniques produce acceptable results in solving the problems of different domains [11]. Nature-inspired algorithms provide better models with fast convergence rates, quicker response times, fair exploration and exploitation, and fewer algorithm-specific control parameters.
There have been various reviews that cover numerous dimension reduction strategies for high-dimensional biomedical data. Most of the previously published surveys focused only on feature selection and extraction algorithms for DNA microarray datasets [12,13,14]. To the best of our knowledge, none of them provide a comprehensive data reduction algorithm, including feature selection, feature extraction, and their hybrid from the standpoint of ML algorithms used. The present review bridges the gap identified above, investigating the severe difficulty of dealing with high-dimensional gene expression datasets and several proposed solutions with nature-inspired algorithms. A taxonomy of data reduction techniques with the two most important processes of data reduction: feature selection and feature extraction, and their hybrid algorithms with nature-inspired techniques, have been defined. Therein, each class is further classified as supervised, unsupervised, or based on the applied machine-learning algorithm. Finally, this review examines frequently used nature-inspired hybrid algorithms and summarizes the deductions and new trends in cancer disease classification and prediction based on high-dimension gene-expression data using machine learning approaches.

1.1. Research Issues

  • What dimensionality reduction techniques are frequently used to manage high-dimensional gene expression datasets?
  • What techniques for feature selection are often used to handle high-dimensional gene expression datasets?
  • What datasets have been frequently used to build the model for cancer classification from high-dimensional gene expression data?
  • Which models have recently been applied to identify the important set of genes needed to correctly classify cancer disease using DNA high-dimension gene expression data?

1.2. Methodology for Review and Article Search and Selection Strategy

The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) review approach has been used to select the appropriate data components for the complete study [15]. Identification, screening, eligibility, and inclusion are the four steps of the PRISMA selection process. Relevant and important keywords were used to find eligible articles, and the actual query string used is shown in Figure 1. This study was limited to studies published in the English language. Most articles published in journals were included, as they provide complete and appropriate scientific work for this survey. We also included other relevant articles to connect certain concepts and techniques. To identify more related papers, we searched and referred to the references of the selected papers. This paper represents a systematic review of dimension reduction techniques for the analysis of high-dimensional gene expression data. This paper also reviewed some of the most popular nature-inspired algorithms with ML frameworks for cancer classification and prediction. We conducted an article search from scientific archives, such as Elsevier, Springer, Wiley, IEEE, Google Scholar, Inder Science, and ACM, covering the years 2017 to 2022. These resources offer a wealth of information about research efforts in a broad but relevant field. Figure 1 completely shows the selection strategy of the final filtered article that is included in the paper.

1.3. Paper Organization

The paper is structured as follows: Section 1 of the paper gives a brief description of different techniques used for the analysis of gene expression datasets. The study then moves on to a comprehensive review of several ML-based feature extraction and feature (gene) selection techniques, including filtering, wrapping, embedding, and hybrid algorithms using supervised and unsupervised approaches in Section 2 and Section 3. The next comprehensive review of widely used nature-inspired algorithms was provided for cancer classification and prediction in Section 4 and Section 5. Section 6 shows the discussion of the review and challenges of nature-inspired algorithms for cancer disease classification and prediction. Section 7 shows the advantages and disadvantages of the reviewed nature-inspired algorithm for cancer classification and prediction. Lastly, the difficulties and unanswered issues surrounding nature-inspired algorithms for accurate cancer classification and prediction were highlighted in Section 8 with a conclusion and future direction.

2. Machine Learning Based Dimensionality Reduction Algorithms

Based on the presence of classifiers in dataset samples, dimensionality reduction strategies are categorized as supervised or unsupervised.
Supervised learning is a sort of ML in which the output is predicted by the machines using well-labeled training data that has been used to train the machines. The term “labeled data” refers to input data that has already been assigned the appropriate output. In supervised learning, the training data that is given to the computers serves as the supervisor, instructing them on how to correctly predict the output. It employs the same idea that a pupil would learn under a teacher’s guidance [16,17,18]. Figure 2 shows the complete taxonomy of dimension reduction algorithms for biomedical data analysis.
Unsupervised learning, this type of machine learning involves no use of tagged classes. This approach hardly ever produces the best subgroups since it ignores the possible relationship between characteristics. It also employs some mathematical concepts without asserting that these concepts are true under all circumstances. In contrast, no prior knowledge is necessary to classify new samples. As a result, it is seen as objective. A subset of the supervised group called the semi-supervised group uses both labeled and unlabeled data. Unlabeled data is typically utilized to determine the geometrical structure of the space, while labeled data is usually used to shorten the distances between data points of different categories [19,20].

3. Types of Dimension Reduction Techniques

The human body is comprised of thousands of genes, although only a few are associated with a cancer diagnosis or complications in the human body. Therefore, for identifying the associated genes in a typical cancer type from any high-dimensional gene expression data, precisely, 2n possible subsets of genes are available for a dataset comprising n features [2,3]. Therefore, a successful dimensionality reduction algorithm provided the minimum number of genes for the learning techniques. A learning approach tends to generalize when it has numerous features, which causes its performance to degrade. There are two methods, that are regularly used to reduce dimensionality: first feature selection and second feature extraction. In order to attain improved learning performance, we required lower computational expenses and significantly improved interpretability. Dimension reduction attempts to choose just a few useful genes from the original set in accordance with such relevance evaluation criteria. Traditionally, the adoption of categorization characteristics seeks to pick the smallest collection of properties that should satisfy categorization accuracy [6]. Figure 3 shows the benefits of dimension reduction methods.

3.1. Feature (Gene) Selection Methods

Feature selection approaches have been split into three classes based on the selection technique: filter, wrapper, and embedded. Without employing any learning models, the feature selection picks the features depending on the data’s statistical properties.

3.1.1. Filter

Filter methods are used to determine the significance of any feature and fall under the univariate and multivariate categories. The univariate filter method first employs a specific criterion to pinpoint the most significant k-feature rankings, after which each attribute is evaluated and given a distinct rating. This type of filter, in contrast to a multivariate method, is unable to spot duplicate features since it ignores feature dependencies that affect classification results. Univariate filters, on the other hand, are quicker than multivariate ones. The most well-known univariate filters employed in several of the evaluations on this list were Fisher Score, Mutual Information, and Laplacian Score (LS). The multivariate filter takes into consideration the relationships between various qualities. As a result, this type of filter has superior classification performance than other types of filters and can evaluate both redundant and irrelevant material. Some of the survey studies utilized the most well-known univariate filters, Relief-f, minimal redundancy, and maximum relevance [21,22].

3.1.2. Wrapper

This feature selection approach relies on classifiers. It operates by choosing a subset of characteristics from a given learning model that yields the best outcomes. Wrappers fall within the stochastic and greedy search strategy categories. A greedy algorithm follows the problem-solving process by making the local optimum choice at each stage. On the other hand, the stochastic search technique is based on the concept of various unpredictable, nature-inspired methods and uses a variety of meta-heuristic algorithms.

3.1.3. Embedded

In an embedded method, feature selection is constructed into the classifier algorithm. During the training step, the classifier adjusts its internal parameters and determines the appropriate feature that produces the best classification accuracy.

3.2. Feature Extraction

Feature extraction is an intelligent substitute for feature selection to reduce the size of high-dimensional data. In the literature, it is also called feature construction or projection onto a low-dimensional subspace. The feature extraction method transforms the original feature into a lower-dimensional space; in this way, the problem is represented in a more discriminating (informative) space that makes further analysis more efficient. There are two main types of feature extraction algorithms: linear and nonlinear. Linear methods are usually faster, more robust, and more interpretable than non-linear methods. On the other hand, non-linear methods can sometimes discover the complicated structures of data where linear methods fail to distinguish [8].

3.3. Hybrid

The hybrid method of dimension reduction combines the benefits of feature extraction and feature selection approaches to improve the classification and prediction accuracy of an ML-based model. This kind of feature selection technique is often used in two phases. Filter techniques, whether a combination of filters or extraction, were used in the first stage [2,3,4,5] to limit the number of features conveyed to the wrapper stage, where classification performance will be improved with a suitable calculation time. They consequently inherit the remarkable computational speed of wrapper algorithms along with the computational effectiveness of filter algorithms. The first stage filter or extraction method has frequently been used to remove redundant genes in the hybrid method with two stages. In the second stage, wrapper algorithms take into account the learning algorithm to choose the biomarker gene subspace from among previously chosen features. The chosen genes were then used to create a prediction model, which was then tested for effectiveness. There have been numerous hybrid-based supervised feature selection methods studied. Hybrid algorithms, which combine filter or extraction with wrapper methods, have been developed in various studies [23,24,25]. Table 1 depicts the advantages and drawbacks of different types of dimension reduction techniques.

4. Nature-Inspired Algorithms

Optimization and its related solving methods are becoming increasingly important in most academic and industrial fields. The goal of the optimization process is to make a system or a design as effective and functional as possible. This has been achieved by optimizing a set of objectives while meeting the system requirements. Optimization techniques are classified into exact and approximate algorithms. Nature-inspired (NI) methods, a sub-class of approximate techniques, are widely recognized for providing efficient approaches for solving a wide variety of real-world optimization problems [26]. Numerous distinct scholars have put out various nature-inspired algorithms over the past few decades. ML classification with nature-inspired algorithms-based data optimization has recently achieved considerable attention in various domains. When compared to other standard optimization techniques, some of these have proven to be remarkably effective [27]. There are numerous ideas out there that a new researcher trying to tackle or solve an issue utilizing nature-inspired methods gets exhausted. Not all techniques are appropriate for every sort of issue; some people excel at others. Nature-inspired methods are a group of cutting-edge techniques and strategies for fixing disputes that are motivated by biological tactics, Particle swarm optimization (PSO), the bat method, ant colony optimization, the cuckoo search algorithm, and the genetic algorithm are different famous examples of optimization algorithms that take their inspiration from nature [7]. The social behavior of biological systems serves as an inspiration for biologically based algorithms. Figure 4 shows the most popular and widely used nature-inspired algorithms.

Characteristics of Nature-Inspired Algorithms (Exploration Phase and Exploitation Phase)

The main phases through which nature-inspired algorithms work are exploration and exploitation. The exploration algorithms search for new solutions in new regions, while exploitation means using already existing solutions and making refinements to them so their fitness will improve. These algorithms may look into many areas of the design space because of exploration’s diversification, boosting the likelihood that they will discover the real global optimality through exploitation. Below are some characteristics of exploration and exploitation [28,29,30].
  • Each individual in a population of different participants, such as particles, ants, bats, cuckoos, fireflies, bees, etc., signifies an initial solution that has been utilized in every method. In terms of objective fitness, the population usually provides the most accurate information.
  • To encourage population development, a variety of operators (such as mutation and crossover) are widely utilized, which are typically stated in relation to computational calculations or algorithms. This kind of growth is usually continuous and produces solutions with a variety of assets. The system is considered to be convergent when all solutions have sufficiently converged.
  • The moves of an agent, which are basically preset, form a piecewise zigzag route in the search space. As a result, strategies for randomization are routinely utilized to provide fresh solution vectors or motions. The algorithm can alter its states (or solutions) thanks to this randomization, allowing it to bypass any local optimum.
  • Every algorithm tries to do some type of local and/or global search. If the search area is mostly local, there is a higher chance of becoming trapped there. If the search focused too much on global movements, convergence would have been impeded.
Through exploitation, the rate of convergence can be accelerated by focusing the search on small areas and utilizing local data from the area. Exploitation can lessen population variety, and strong local leadership can even increase the consistency of solutions. Even if neither is optimal, both extremes may prevent an algorithm from being convergent because they lessen the possibility of actual global solutions. The conclusion is that a careful balance of exploration and exploitation is required for every nature-inspired algorithm, and this balancing act may be determined by algorithmic design and difficulty [9]. Figure 5 shows the working procedure of nature-inspired algorithms in the exploration and exploitation processes.

5. Review of Nature-Inspired Algorithms for Dimension Reduction of Gene Expression Data

ML classification with nature-inspired algorithms-based data optimization has recently achieved considerable attention in genomic research. In this paper, a number of selected nature-inspired algorithms are systematically reviewed and analyzed for gene selection. This paper mainly focused on the original principles behind each of the algorithms and discussed their applications in gene selection for cancer classification and prediction. Table 2 shows the most popular nature-inspired algorithm used for cancer classification and prediction.

5.1. Cuckoo Search Algorithm (CSA)

The behavior of cuckoo birds during reproduction serves as the basis for CSA optimization techniques. The cuckoo lays one or more eggs in the nests of other birds in an effort to maintain the continuity of their generation by encouraging the host birds to follow their innate instincts to reproduce, hatch, and feed the young cuckoos [3,4,5,6]. Cuckoo eggs typically hatch first before host eggs, but certain such eggs can even duplicate or resemble the nest of the feeding bird, and some of them even have the potential to develop and have more access to food than the host birds. Such eggs will mature into complete cuckoos if the host birds do not recognize them. The two sorts of cuckoos used in this simulation are eggs and adult birds. Eventually, the remaining eggs will develop into mature cuckoos, which will then congregate to form a community. The optimal breeding and reproduction conditions are determined by using the CS algorithm, which finds the global maximum of objective functions.
Working rules of CSA:
  • A cuckoo has never laid more than one egg in a random nest.
  • The generation that comes after will inherit the best nest and eggs.
  • The host bird’s pa is between 0 and 1, which is the chance of finding a cuckoo egg given a certain number of host nests.
According to these 3 considerations. The host bird might either leave the nest and start a new one or abandon the egg. The last premise can be approximated as the ratio pa of the n nests replaced by new nests.
Application of Cuckoo Search algorithm for biomedical data.
Prabukumar et al. provided an ideal approach to increase accuracy in the Fog computing environment to detect lung cancer. Fuzzy C-Means and the region-growing segmentation algorithm were applied to precisely segment the regional interests. The important properties of the investigated nodule, including its geometrical, textural, statistical, and intensity traits, have then been retrieved. The optimal qualifications for identifying pulmonary cancer are determined based on the above qualities using the cuckoo search optimization approach [30].
Alzaqebah et al. applied a cuckoo search approach for the attribute selection process that stores the most advantageous features discovered by combining the most effective techniques with a memory-based method. At each iteration of the new technique, identify the characteristics of chosen features that increase classification precision. Using microarray datasets, the original algorithm and the modified technique have been compared, and it has been shown that the modified technique yields adequate results when compared to the original approaches [31].
In order to classify cancer, Gunavathi et al. employ the CSA approach to choose criteria using data from microarrays on gene expression. In this study, significant genes were first identified using CSA as a feature selection method, and then the identified genes were ranked to improve classification accuracy. To rank the important genes, T-statistics, SNR, and F-statistics values were used. To find informative genes, the CS of the highest m-rated genes is employed. The fitness function for CS is the classification accuracy of the k-nearest neighbor classifier. The data shows that CS provided an average accuracy of 100% with DLBCL cancer data [32].
Arjmand et al. presented a swarm strategy for an autonomous segmentation method in MRI data. They used the cuckoo search optimization approach to localize data using the K-Means algorithm’s centroids and K-Means clustering. According to results from the RIDER breast dataset, the recommended technique outperforms alternative strategies such as the conventional K-Means clustering algorithm and Fuzzy C-Means [33].
Sampathkumar et al. showed how the cuckoo search, which is influenced by biological processes, is capable of sorting many cancer subtypes with extraordinarily high accuracy while selecting genes from microarray technologies. Five benchmark datasets for the inquiry’s findings were compiled by utilizing cancer gene expression. According to the research, modified cuckoo search performs better than CSA and other well-known approaches. Prostate, lung, and lymphoma have been accurately diagnosed with 99% accuracy for the top 200 genes in the dataset. Modified cuckoo search for the colon dataset is 98.54%, compared to 96.98% for the leukemia dataset [34].

5.2. Bat Algorithm

Bats are amazing species that employ a powerful audio transportation system called bio-sonar. In the dark, animals employ bio-sonar, widely known as echolocation, to search for food and/or travel to avoid hazards. In order to hear the echoes that are radiated from smaller objects, bats emit a high sound frequency. The frequency is connected to their ways of getting food. Short-intensity signals are employed by bats to travel around in an interval. The overtime usage of bats varies by species [35]. Microbats’ echolocation features show a few basic or idealized guidelines that can be used to improve the initial bat algorithm or develop a brand-new algorithm that is influenced by bats. Loudness and pulse emission rates must be adjusted as iterations advance. The volume frequently drops off, and the rate of pulse production quickens as the bat approaches its target.
Working rule of Bat Algorithm:
  • Bats use echolocation to determine distance. In some amazing ways, they can recognize the distances or intervals between the surrounding obstacles and the prey.
  • In search of food, bats fly freely at various velocities and frequencies with varying wavelengths and loudness. They may continuously modify the wavelength and rate of release of loudness of the flashes based on how close their objective.
  • Furthermore, loudness could differ in a range of ways. In this case, it is presumed that the volume fluctuates between a high and a low fixed value.
Application of the Bat algorithm for biomedical data.
Betar et al. suggested a mixed wrapper/filter as a solution to the issue of gene selection. This method has enhanced the bat algorithm and is resistant to very little redundancy. The most promising genes were discovered using the maximum relevancy and minimum redundancy filter method. The engine in wrapper method was suggested for identifying a select few important genes. With the use of 10 gene expression datasets, the effectiveness of the suggested method was evaluated. Enhanced Bat algorithm performance has been evaluated by contrasting the convergence behavior of Teoriya Resheniya Izobretatelskikh Zadatch (TRIZ) optimization operators with and without them. Ten cutting-edge datasets were compared for evaluation against the specified rMRMR-results MBAs. According to the comparison study, the recommended approach showed positive outcomes in two of the three situations in terms of categorization precision and gene selection [36].
Jeyasingh et al. created the Modified Bat Algorithm (MBA) for feature selection in order to filter out unnecessary traits from a beginning dataset. To choose random examples from the data, simple random sampling was used to upgrade the Bat technique. The main features of the dataset were ranked using the best features made available globally. A classification technique named Random Forest (RF) was trained using the chosen features. The MBA feature selection technique increased the RF’s classification accuracy for detecting the presence of breast cancer. The effectiveness of the suggested MBA feature selection method was examined using the Wisconsin Diagnostic Cancer Dataset (WDBC) [37].
Alhassan et al. created a technique for autonomously segmenting objects that makes it possible to distinguish cancerous lesions from MRI images. The ECN algorithm evaluated the suggested Enhanced Capsule Networks (ECN) classifier based on the measurable properties of accuracy, precision, recall, and F1-score over MRI images. A genetic algorithm was used in the automated tumor stage categorization, which enhanced the classification accuracy [38].
Hambali et al. proposed an approach for identifying appropriate and instructive features from highly directed microarray cancer data sets. To assess the method, the four classifiers Decision Tree, Random Forest, C4.5, and Regression Tree were used. The outcomes illustrated how the technique may be applied to categorize microarray cancer data. In all seven test data sets, the random forest achieves 100% accuracy with only a few genes [39].
Chatra et al. proposed a technique for categorizing cancer using gene expression data. They also recommend a fresh fitness feature to improve the binary bat method’s feature selection. The first fitness measure identified in the literature has been contrasted with the suggested methods. According to the results of the tests that had been performed, the latter is less effective than the former [40].

5.3. Genetic Algorithm (GA)

A number of algorithms that focus on the concepts of genetics and natural selection are identified as genetic methodologies. The genetic algorithm employs a statistical transition method, whereas existing techniques utilize gradient knowledge. These qualities of the genetic algorithm enable it to serve as a general-purpose optimization method. They are utilized in a number of optimization problems, function approximation, and machine learning applications because they also enable ways to investigate irregular space [41,42,43]. A precise or approximative resolution to an optimization or research issue can be obtained using a genetic algorithm. Genetic algorithms, a subclass of simulated annealing, use mechanisms like mutation, selection, and crossover that are based on biological evolution. They can utilize algorithms to replicate annealing, hill climbing, or tattoo search. Genetic algorithms are a subset of evolutionary algorithms. The use of evolutionary algorithms facilitates the resolution of problems for which a clear, effective answer is still lacking. Using evolutionary algorithms, scheduling, shortest path, and other optimization problems have been resolved. In very many domains, particularly biology, engineering, computer science, and social science, evolutionary computation has been utilized to solve complex situations in the most efficient way possible. In the domain of bioinformatics, GA frequently uses algorithms for different purposes [44].
Working rule of GA:
The genetic algorithm is a technique for solving both constrained and unconstrained optimization problems that are based on natural selection, the process that drives biological evolution. The genetic algorithm repeatedly modifies a population of individual solutions. GA works with ideas of mutation, crossover, and selection using the terminology of genetics (from the current generation to create the next population), and as a result, the algorithms become more effective. Five phases are considered in a genetic algorithm.
i.
Initial population: This algorithm starts with the selection of initial sets, which may or may not include the optimal values. These sets of values are called ‘chromosomes’ and the step is called ‘initialize population’.
ii.
Fitness function: The value of the objective function for each chromosome has been computed with some fitness value in this step.
iii.
Selection: This step is very important and is called ‘selection’ because the fittest chromosomes are selected from the population for subsequent operations.
iv.
Crossover: This step is called crossover because, in this step, chromosomes are expressed in terms of genes.
v.
Mutation: Mutation is the process of altering the value of a gene to find the best solution.
At various times, careful juggling between the level of selection and the appropriate ratio of crossover and mutation is required for finding the best optimal solution. Figure 6 shows the working procedure with various phases of GA [45].
Application of Genetic algorithms for biomedical data.
For the gene selection challenge, Zexuan et al. developed a novel Markov blanket-embedded genetic algorithm. The GA algorithm is swiftly improved and fine-tuned by the embedded Markov blanket-based memetic operators. Markov blanket and predictive power in classifier models both give empirical findings. When modified GA was evaluated on benchmark microarray datasets, it successfully and effectively removed redundant and superfluous genes to improve classification accuracy [44].
Motieghader et al. proposed a hybrid meta-heuristic method that combines genetic algorithms with learning circuits. In this paper, a hybrid meta-heuristic algorithm, called GALA (Genetic Algorithm and Learning Automata), is proposed for cancer classification. Six different cancer datasets, including Colon, ALL AML, SRBCT, MLL, Tumors 9, and Tumors 11, were chosen to evaluate the performance of GALA. Based on the evaluation method, the GALA algorithm outperformed several recently proposed algorithms on the same dataset [45].
Jansi et al. suggested a novel method for gene selection by utilizing a two-stage feature selection technique with mutual information and GA to find pertinent genes for the classification of cancer data. Genes that have a high mutual information value were included in the second stage. Next, a genetic algorithm was used to choose the appropriate group of genes needed for precise identification. For categorization, a support vector machine is employed. The result shows that for datasets of ovarian, lung, and colon cancer, mutual information with GA as a gene selection strategy surpasses conventional methods in terms of classification accuracy [46].
The innovative GARF (Genetic Algorithm Based on the Random Forest) approach was created by Paul et al. to characterize clinical imaging data. The proposed algorithm was tested on 65 patients with mildly advanced esophageal cancer who were candidates for chemoradiation therapy. The most important characteristics that predict treatment response or prognosis three years after treatment were identified using GARF. The random forest misclassification rate was 18% for the set of nine variables that accurately predicted results. The maximum area under the receiver operating characteristic (ROC) curves was 0.823. In both prognostic and predictive evaluations, GARF performed better than the other four methods [47].
Lu et al. proposed a unique feature selection technique by utilizing the GA’s specially crafted trace-based separability criterion. Without respect to any particular categorization, the proposed parameter assesses the relevance of the feature subset using the subclass specificity and variable separability scores. A common lung cancer dataset was used in the study. Using three classifiers (the support vector machine, the back-propagation neural network, and the K-nearest neighbor), we compared the results of the proposed techniques to those of other popular feature selection techniques. The comparison results show that the proposed intelligent system for early lung cancer detection and diagnosis problems was successful [48].

5.4. Whale Optimization Algorithm (WOA)

WOA is an innovative meta-heuristic approach that gets inspiration from nature and replicates the teamwork of whale sharks in chasing their food. The bubble-net approach is used by whale sharks, in which two or maybe more whales swim in a tight circle while blowing bubbles to surround a group of prey along a spiraling course. While other whales or hunters try to force the prey into the net, others try to drag it towards the surface [49,50].
Working rule of WOA:
The humpbacks utilize the following behaviors, which include bubble-net attacking and hunting for prey. When using a bubble-net attack, humpback whales may really locate their target and approach it in a diminishing circle. WOA believes that the best possible choice solution at the time was the prey or very close to the prey in the optimal solution, since the best solution was not initially known. Its other search agents try to place them in the best search agent’s direction. While exploiting might well be achieved by using the leading search agent, as in the bubble-net methodology, exploration can be achieved by using a random search agent [50,51,52].
Application of the Whale algorithm for biomedical data.
Abbas et al. developed a ground-breaking technique for effectively choosing features by combining the whale optimization with the exceptional randomness approach. The WOA minimizes the dataset dimension and gets rid of the elements that are required for accurate classification. Experimental results demonstrated enhanced performance of eight popular ML algorithms, including Support Vector Machine, Kernel Support Vector Machine, Decision Tree, Logistic Regression, Stochastic Gradient Descent, Gaussian Naive Bayes, and k-Nearest Neighbor. SVM came in first place with a 98.60% accuracy rate with the proposed algorithm of gene selection. Experimental data also shows how feature selection strategies may raise prediction precision [49].
Fang et al. suggested a whale optimization strategy-based technique for the MLP network. The proposed MLP network has the potential to handle a variety of issues with the whale optimization strategy-based technique. The Mammographic Image Analysis Society database, which comprises 322 digital mammography images, and the Digital Database for Screening Mammography, which has roughly 2500 digital mammography images, were used to validate the suggested technique. The proper detection rate, percentage of identification with false acceptance, and percentage of identification with false rejection were analyzed and compared using different approaches in terms of detection. The outcomes demonstrate how much more accurate and efficient the suggested technique is compared to alternatives [50].
Stephan et al. combined WOA’s bubble net assaulting technique with ABC’s exploitative employee bee phase for the hybrid frame model of cancer classification with an artificial bee colony and WOA. During the assaulting phase, honeybees use humpback whales to find more advantageous areas for food supplies. The inadequate investigation of ordinary ABC is strengthened by the suggested mutative initialization step, which makes up the explorative stage of the proposed algorithm. An ANN model’s parameters have been optimized using the proposed approach. Levenberg-Marquart, robust back propagation, and momentum-based gradient descent are among the backpropagation learning techniques used compared to the proposed model of MLP. Several breast cancer datasets have been used to assess the precision, complexity, and processing time of the proposed model [51].
A hybrid intelligent model presented by Sayed et al. for the study of clinical breast cancer data that combines feature selection tactics with bio-inspired cluster analytical methodologies. Both the insect flame and whale optimization methods have binary variants. The presented methodologies were assessed using various assessment criteria. The experimental results demonstrated that the recommended bio-inspired feature selection approaches may offer useful feature subsets and data partitions using swarm measures and figure measurements [52].
Sayed et al. used the strongest point subset from the Wisconsin Breast Cancer Database (WBCD) for classification with the new proposed meta-heuristic optimization technique. The latest optimization techniques are based on the bio-inspired architecture of WOA. A number of criteria, including precision, accuracy, recall, and f-measure, were evaluated on WBCD data using the new proposed technique. The outcomes have been examined and contrasted with other breast cancer detection algorithms. The results of the trial show how good the WOA algorithm is at identifying breast cancer. Genetic algorithms, principal component analysis, mutual information, statistical dependencies, random subset feature selections, sequential forward selections, and sequential floating forward selections have all been compared to it. In total, the overall accuracy, precision, recall, and f-score for the proposed method were 98.77%, 99.15%, 98.64%, and 98.9%, respectively [53].

5.5. Harris Hawk Optimization (HHO) Algorithm

Heidari et al. created the HHA method as a labeling algorithm inspired by nature and matching the characteristics of Harris hawks. Harris hawk is recognized as an illustration of the most intelligent species, depending on its own food patterns. It is a species of bird of prey that lives in families and is indigenous to the southern Phoenix area of the United States. Animal researchers have found that the Harris hawk actively cooperates with its group. In other words, the majority of other raptor species attack and consume prey entirely on their own. Each member of a Harris hawk group participates in the evening meal, which increases all through the non-breeding months [54,55,56].
Working rule of HHO:
The Harris hawks use some really amazing behavioral patterns to identify and bite their prey. Harris hawks capture prey with such a sudden pounce approach. The main objective of this technique, commonly known as “Seven kills,” is to weaken the prey before trapping it. A number of actions must always be taken in order for the operation to be productive. A first step is to attack fast (within a few seconds) in an attempt to startle the victim. Then, over the course of several minutes, the group members alternately engaged in a series of similar, quick dives at the prey to confuse it. Following that, an experienced and skilled hawk effortlessly grabbed the prey. The team finally shared the prey with the other members. The exploration and exploitative portions of the algorithm are conducted here by looking for food, delivering a quick strike, and employing a number of attacking techniques [56,57].
Application of the Harris Hawk algorithm for biomedical data.
Kaur et al. developed DLHO by combing Dimensionality Learning-based Hunting (DLH) with HHO for biological datasets. This method was created by fusing HHO with the elements of the DLH to address the issues of HHO like the imbalance between exploration and exploitation, the lack of population variety, and the early convergence of the HHO. The DLH search technique takes a new tack in an effort to paradigm a neighborhood for each search member and allows search agents to share neighborhood information. This method aids in maintaining variety and proportionality between local and worldwide searches [54].
Rammurthy et al. located tumor cells using MR images with the newly created Whale Harris Hawks Optimization (WHHO) technique. In this technique, data segmentation has been performed using cellular automata and rough set theory. Additionally, local optically oriented pattern, mean, variance, and kurtosis were evaluated for tumor volume. A proposed method is used to train deep convolutional neural networks for brain cancer detection. The suggested WHHO-based deep convolutional neural networks outperformed previously suggested methods with maximum accuracy, specificity, and sensitivity [55].
Arif et al. created an autonomous system for classifying skin lesions using transfer learning and deep learning network. To identify melanoma in its early phases, the digital dermoscopy images of the skin were initially processed using image processing techniques. Next, these images were then post-processed for image enhancement and pre-processed for noise reduction to build a detection system. The picture has been subjected to updated K-Means clustering and K-Means image segmentation for feature extraction and first-order statistics, and characteristics have been retrieved. Finally, several classifiers have been employed in order to forecast the phases and effectiveness of the suggested task using HHO’s chosen features. The sensitivity, precision, accuracy, and specificity metrics were used to evaluate the device. With high classification rates, the recommended technique outperformed the output of current methodologies [57].
To identify breast cancer, Jiang et al. suggested a hybrid model based on two optimizations strategies and machine learning. Before differentiating between benign and malignant samples, they first used feature weighting based on K-Means to solve the issue of nonlinear and uneven data distribution. Then, Harris Hawks Optimization is combined with PSO as a new proposed method, PHHO. Furthermore, the extreme learning machine has been optimized using PHHO. The Wisconsin diagnosis breast cancer data set is then used to check the proposed model. The findings indicate that the suggested model may achieve 98.76% accuracy, 97.37% sensitivity, and 99.46% specificity [58].

5.6. Ant Colony Optimization (ACO) Algorithm

The first ACO algorithms were released by Marco Dorigo and associates in the early 1990s. These algorithms were developed with inspiration from research on ant colonies. Ants are social insects. They are social animals that live in colonies, and the survival of the colony rather than the individual determines how they will act. The foraging techniques of ants, particularly their skill at figuring out the quickest routes between food sources and their colony, served as an inspiration for ACO. It has been demonstrated that ants employ stigmergy, an acoustic system that uses pheromone trails as a medium of indirect communication, to determine the quickest routes between their colony and food sources [59,60].
Working rule of ACO:
One of the newest methods for approximation optimization is ACO, which is based on real ant colonies. Ants leave a pheromone chemical trail on the ground as they move across the area near their nest when they first start looking around randomly for food. They frequently choose paths with high pheromone concentrations while deciding where to go. After finding a source, an ant looks at the quantity and quality of the food and gives some of it back to the nest. The size and quality of the meal may have an impact on how much ground-level pheromone and deposits are made on the animal’s return trip. Other ants will follow the pheromone trails to the food source. Ants’ feeding habits have an influence on ACO. Indirect communication between ants employing chemical trails is the basis of this activity, which enables them to locate an efficient path between their colony and food sources. For instance, ACO algorithms use this trait of actual ant colonies to handle discrete optimization issues [61,62].
Application of the Ant Colony algorithm for biomedical data.
Sun et al. presented a potent mongrel gene selection strategy for cancer classification based on the Relief-f and ACO. The Relief-f approach first inserts the average distance between the k-nearest or k farthest neighbor samples. It also performs better when gene weight values are predicted for data based on the distances between samples belonging to the same class or to other classes. In order to transmit reliable results under pressure and simplify the weight measurement of genes, a distance measure was developed. The outcomes of the trials using a variety of publicly available gene expression datasets demonstrated good classification accuracy with the suggested technique. The proposed technique reduces the dimensionality of gene expression datasets by choosing the most pertinent genes [59].
Sumeyya et al. proposed a classification algorithm to categorize the various forms of skin cancer. This study described a hybrid method that combines an ACO and a support vector machine with a Gaussian radial basis function. With the help of 10-fold cross-validation, the suggested hybrid algorithm has been examined with two different data sets of skin cancer. The results of the proposed algorithm on two taken data sets demonstrated 98% and 97% classification accuracy of the support vector machine classifier with a processing time of 26.5 and 11.9 s, respectively [60].
In order to identify the ideal Raman characteristics, Fallahzadeh et al. analyzed 49 spectra from breast tissue samples that were benign, cancerous, and healthy utilizing a 785-nm micro-Raman device. The objective was to increase the accuracy of cancer diagnosis using Raman technology. The biological sample (“12 primary Raman bands”) intensities were retrieved as characteristics of each spectrum after preprocessing to lower background fluorescence and noise. The suitable genes for the diagnosis were then selected using the ACO algorithm. The outcomes showed that when using five characteristics, the categorization accuracy of the normal, benign, and malignant categories increased by 14% and reached 87.7%. The diagnostic accuracy of Raman-based diagnostic models was increased by utilizing ACO feature selection [61].
To reduce mortality, Zainuddin et al. proposed an early diagnosis framework for cancer by using the ACO feature selection method. In the proposed framework, ACO is combined with an adaptive network-based fuzzy inference system. Colon and lung cancer data have been taken from the Kent Ridge Biomedical Data Set Repository to test the proposed model. The number of selected genes based on the strategy of ants has an impact on the classification rate as well as the accuracy of the results. For lung and colon cancer, the highest accuracy rates are 100% and 94.73%, respectively [62].
Rajagopal et al. presented feature optimization approaches; a technique was proposed for diagnosing glioma brain tumors based on random forest classifiers. An ant colony optimization technique is currently being used to improve the feature sets provided by brain MRI scans, which include textural qualities. The random forest classification approach is used to train and classify these optimal sets of features. Based on an optimal collection of characteristics, this classifier separates the brain MRI image into groups for gliomas and other conditions. Additionally, to remove the tumor sections from the glioma image, an energy-based segmentation approach has been used. According to this study’s findings, the suggested approach’s sensitivity, specificity, and accuracy for glioma brain tumors are 97.7%, 96.5%, and 98.01%, respectively [63].

5.7. Artificial Bee Colony (ABC)

The ABC, first proposed by Derviş Karaboa in 2005, is an optimization method based on the intelligent foraging behavior of honey bee swarms. The colony under the ABC model is made up of three types of bees: workers, observers, and scouts. One artificial bee is thought to be present at a time for each food source. In other words, the colony has the same number of working bees as there are food sources nearby. Working bees travel to their food supply, return to the hive, and dance around this location. The working bee, whose food supply has been lost, turns scout and begins looking for a new food source. Bees in the workforce perform dances for onlookers, who choose food sources based on the dances. The following list contains the algorithm’s key steps [2,3,4,5].
Working rule of ABC:
All employed bees receive fresh food supplies. Each worker bee visits a food source in her memory to find the nearest one. Next, it calculates how much honey it has and dances inside the hive. Each observer sees a single bee perform a dance, then, based on the dances, selects one of their sources and goes there. Finally, she selects a nearby neighbor and then considers how much honey it contains. Food sources that are abandoned have been identified and replaced with fresh sources that scouts have found. The ABC algorithm works with three classes of bees. Figure 7 shows the working phase of the ABC algorithm [64,65,66].
Employed Bees (EB): In the colony’s first part, just a few SB and OB are present. OB waits near the hive to be recruited. SB is assigned to look for a prospective nectar source while. Any SB that discovers a nectar supply will transform back into EB. After returning to the hive with some nectar, EB dances in various shapes to inform OB of the source’s information. Diverse dance forms depict vastly different nectar’s supply qualities.
Onlooker Bees (OB): Each OB assigns a quality rating to the nectar sources that were discovered by all EB, after which it moves from one EB to the supply that matches. According to some probability, all OB choose EB. Higher sources (with more nectar) are more enticing to OB and are therefore more likely to be chosen.
Scouts Bee (SB): Once any sources are depleted, the related EB may discontinue using them, convert them to SB, and look for alternative sources. Limit and scout production as part of the abandonment criterion for producing an initial food source. The supply linked to this counter is assumed to have been used up and abandoned if the counter value exceeds the “limit,” or management parameter, of the ABC algorithmic rule. The scout replaces the meal supply left behind with a fresh supply that symbolizes the organization of essentials’ feedback mechanism and fluctuation feature.
Application of the Artificial Bee Colony algorithm for biomedical data.
Coleto et al. developed a hybrid approach to the issue of picking the smallest subset of the important genes for disease categorization. This novel method combines cistron filtering as a first step and an improvement rule as a second step. The most crucial genes from the dataset are selected in the first stage using the Analytic Hierarchy approach, which combines five ranking algorithms. The number of genes that must be controlled throughout this process is reduced because of cistron filtering. Reducing the number of chosen genes and improving classification accuracy were the two objectives of the second step’s cistron selection. In stage two, an ABC-supported dominance rule has been anticipated. A variety of multi-objective strategies that have been suggested in the academic literature are compared with the outcomes of this methodology, which was evaluated using 11 actual cancer datasets. The classification performance with a small set of genes produced by the proposed technique gave astounding results [64].
R. M. Aziz proposed the unique nature-inspired algorithm-based framework for gene selection and cancer classification. She employed CSA to enhance the performance of the ABC algorithm in the onlooker bee phase. Independent component analysis has been performed to reduce the dimension of gene expression data for Naïve Bayes classification in the first step. In the second step, modified ABC was used to optimize the extracted gene. The proposed framework’s effectiveness is demonstrated by the improved performance of the Naïve Bayes classifier with a smaller number of selected genes [65].
Penitha et al.’s hybrid artificial bee colony with a whale optimization combines ABC’s exploitative worker bee component with whale optimization’s bubble web offensive strategy. In the first stage, worker bees use humpback whales to locate areas with more food supplies. The proposed algorithm improved the poor exploration of standard ABC by using whale optimization. In this paper for classification, the proposed algorithm rule has been applied to the ANN model’s variable modification that enforces mistreatment backpropagation learning and includes momentum-based gradient descent, Levenberg-Marquart, and resilient backpropagation. This hybrid model showed improvements in accuracy, complexity, and method time assessed using numerous carcinoma datasets compared to other popular models of ANN [66].
Karthiga et al. used the LUNA 16 dataset as an input dataset for noise reduction and picture improvement for early lung cancer detection with Naïve Bayes classifier. The authors also proposed a novel evolutionary algorithm named Accelerated Wrapper-based Binary Artificial Bee Colony (AWB-ABC for efficient gene selection to improve the performance of the Naive Bayes classifier for classifying the different stages of lung cancer. Firstly, they were pre-processed with noise removal and image enhancement using MMSE. Next, with the recovered picture, the AWB-ABC algorithm was utilized to choose the important genes that have the greatest risk of developing cancer. In the last stage of the proposed algorithm, we effectively categorized LUNA 16 with an improved Naive Bayes classifier. The outcomes show that the recommended strategy outperformes the frequently employed techniques. The recommended method improves classification accuracy by 98% [67].
To improve investigative performance and reduce unhealthy mole fatalities, Aljanabi et al. contracted the cataloguing of photographs using data from multiresolution analysis and bee colony technology. This methodology system intends to improve some of the current approaches and offer novel approaches for the precise, efficient, and dependable automated evaluation of skin lesions. To classify skin cancer, this data is subsequently fed into several well-known algorithms. This approach enables the segmentation process to be used to improve data management, develop security standards, and reduce the likelihood of skin cancer lesions. One of the most crucial processes in dermoscopy image processing is melanoma segmentation. The test results showed that the suggested technique outperformed baseline photographs of skin cancer lesions validated by dermatologists. According to research on databases of skin lesions, the bee colony often has the greatest amount of improvement. The plan’s elements were calculated with care, accuracy, and precision [68].

5.8. Firefly Algorithm (FFA)

Numerous challenging optimization issues have been successfully resolved Using algorithms for swarm intelligence, the subject of extensive research. One of the most advanced and successful swarm intelligence methods and optimization algorithms is Yang’s firefly algorithm. The firefly method has been used extensively and successfully to tackle several optimization issues [69].
Working rule of FFA:
The firefly algorithm was developed using a model of firefly social behavior based on flashing lights or other forms of flash attraction. By simulating the attraction of the flashing features of fireflies, it is possible to simulate how fireflies interact with these flashing lights or other fireflies. Each member of the FFA is represented by a firefly in a swarm. A signaling method that draws more fireflies in is the firefly flash. In the dimensional search space, each firefly stands for a particular strategy. The more promising locations are thought to be in the brighter areas. The computer then tries to assist the fireflies in finding these spots inside the search region. The brightness of a firefly is related to the objective function of the optimization problem and determines how enticing it is. As a firefly moves away from its intended target, its brightness lessens. They are drawn to the firefly because of the variations in brightness between them. This implies that a less luminous firefly may draw in a more brilliant firefly. A solitary firefly will fly anywhere it pleases if none of the others are any brighter than it. Throughout the search process, fireflies might wander to different locations or positions because of their innate attraction, which aids them in finding new possible optimal solutions [70,71,72,73].
Application of the Firefly algorithm for biomedical data.
A metaheuristic continuous rule with the firefly approach has been developed by Thanoon et al. to figure out the PSVM’s carangid penalty’s calibration parameter. This framework has great classification ability and can identify the most important genes. The proposed rule performed better than other competitive methodologies for four benchmark datasets of organic phenomena in terms of classification accuracy and fewer selected genes. This is demonstrated by experimental findings using these datasets [69].
Sawhney et al. used a random forest classifier with FFA for the diagnosis of breast, cervical, and hepatocellular carcinomas. The proposed hybrid model has demonstrated an improvement in classification accuracy and gene reduction when compared to other contemporary techniques. They suggested using the Binary FFA to significantly condense the feature set to an optimal subset. They also added a punishment component to the current fitness function for the improvement of gene selection [70].
Farouk et al. used a novel optimization method to choose fewest number of important genes for the prediction of breast cancer. Two optimization methods, FFA and PSO, were combined in a proposed framework. To determine the significance of particular subsets of potential-generating genes, crude pure mathematics is used. The projected rule was assessed and contrasted with the approximate, established ensemble reduction rule of the rough set and proposed algorithms with breast cancer datasets. The statistical findings reveal that the suggested hybrid rule has a high degree of accuracy in cancer categorization. Ackley, Griewank, Levy, and many more benchmark fitness functions are included in the intended rule to compare experimental outcomes. The hybrid framework rule offers high-accuracy cancer dataset categorization using the top benchmark functions [71].
Alshamlan et al. report that for the investigation of microarray cancer organic phenomena, the SVM classifier and the FFA feature selection methodology have been developed and named FFF-SVM, a new biomarker cistron identification algorithmic method. The classification accuracy for the selected cistron set was evaluated using the leave-one-out cross-validation (LOOCV) technique and an SVM classifier. Five benchmark binary and multi-class microarray datasets were used to test the FFF-SVM rule. They contrast the anticipated outcome with other connected, evolving algorithms to assure correctness. The experiment demonstrated that, with its high classification accuracy and random distribution of elite genes, the FFF-SVM outperformed another hybrid framework. The desired FF-SVM wrapper-based factor choice rule is contrasted with the predicted rule. According to the findings, the hybrid-based framework performs better than the other popular wrapper-based framework. The actual regulation [72] was better than we expected.
Nguyen et al. suggested a technique for diagnosing cardiopathy by utilizing rough sets supported by attribute reduction with type-2 phase datasets. High-dimensional issues in biomedical datasets and uncertainties have been resolved with clustering-based attribute reduction. As part of its hybrid learning strategy, chaotic FFA and GA with fuzzy c-mean cluster rules named IT2FLS have been applied to the dataset. The proposed learning approach for a high-dimensional dataset has a significant procedural cost. The chaotic FFA approach yields the most effective approximate sets-based attribute reduction, making the IT2FLS a more significant method of gene reduction. The experiments’ findings demonstrated that the suggested approach outperforms old-fashioned ML methods like artificial neural networks, support vector machines, and Naive Bayes. The suggested model gave a good gene selection network for identifying cardiac issues [73].
Banu et al. suggested an inexpensive fuzzy-firefly clump by using the properties of a firefly and a fuzzy clump. A comparison is made between the proposed technique and other swarm optimization-based clump methods. It is applied to five datasets of organic phenomena. The proposed algorithmic approach generates clusters that provide interpretations of various organic phenomenon patterns found in cancer datasets. Experimental findings showed the fuzzy-firefly clump’s remarkable ability to discriminate between genes that are co-expressed and co-regulated [74].
Almugren et al. introduced a new wrapper feature selection technique called FF-SVM for identifying cancer from the microarray. To evaluate the classification rate, FFA was applied along with SVM classification with leave-one-out cross-validation. Five benchmark binary and multi-category microarray datasets have been used to assess the FF-SVM formula. They compared the proposed algorithm with different connected progressive algorithms to confirm the outcomes. The results showed FF-SVM has excellent classification accuracy even when using only a limited number of manually chosen genes [75].

5.9. Particle Swarm Optimization Algorithm

PSO is a stochastic optimization technique based on the movement and intelligence of swarms. PSO, one of the microbiological approaches, is unambiguous in its search for the best solution to the problem at hand. PSO was first developed as a stylized representation of the movement of creatures in a bird flock or fish school to model social behavior. PSO is a metaheuristic in the sense that it can search very large areas for potential solutions and makes little to no assumptions about the issue being optimized with a few hyper-parameters. Furthermore, unlike traditional optimization techniques like gradient descent and quasi-newton methods, PSO does not employ the gradient of the issue being improved, negating the need for the optimization problem to be differentiable [76,77,78].
Working rule of PSO:
The core PSO method employs a swarm of potential solutions (called particles). Using a few straight-forward equations, these elements are moved around in the search space. The positions of both the individual particles and the entire swarm that are best known serve as a guide for the particle movements. These will eventually start to direct the swarm’s motions once better sites are found. In PSO, the concept of social interaction is used to solve a problem:
  • It uses a number of particles (agents) that constitute a swarm moving around in the search space, looking for the best solution.
  • Each particle in the swarm looks for its positional coordinates in the solution space, which are associated with the best solution that has been achieved so far by that particle. It is known as “local best”.
  • Another best value known as gbest, or global best, is tracked by the PSO. This is the best possible value obtained so far by any particle in the neighborhood of that particle. Figure 8 shows the process of the particle swarm optimization algorithm [79,80,81].
Application of the Particle Swarm algorithm for biomedical data.
Hu et al. proposed a novel federated feature selection technique based on the PSO algorithm. In this framework of feature selection inspired by the idea of federated learning, a credible third participant was used to process and integrate optimal feature subsets. The proposed framework with PSO effectively solves feature reduction problems with multiple participants. To improve the ability of the proposed algorithm, compared with several typical assembling feature selection algorithms on 15 data sets, experimental results showed that the proposed algorithm can significantly improve the classification accuracy of the feature subset selected by each participant [77].
Thanoon et al. applied a penalized support vector for the cancer classification. In this work, they proposed a hybrid metaheuristic algorithm by using the firefly algorithm and PSO for tuning the parameter of the penalized support vector. The proposed algorithm efficiently exploits the strengths of both the firefly and PSO algorithms in finding the small subset of relevant genes needed to increase the classification performance of penalized support vectors. When compared to other competitor techniques, the proposed hybrid combination of gene selection outperforms them in terms of classification accuracy of penalized support vectors with the fewest gene [78].
PSO and an adaptive K-nearest neighbor (KNN)-based sequence selection method were suggested by Kar et al. as a way to locate a limited number of beneficial genes that are appropriate for the classification purpose. The proposed framework for deciding how many neighborhoods should be evaluated, to categorize the data efficiently with the precise number of K. As a result, in order to select the ideal values of K, a heuristic that is effective and guided by classification accuracy is projected. This technique identified the smallest feasible collection of genes that accomplish a certain task using three benchmark microarray datasets. The findings showed how effective the suggested method is in terms of classification accuracy on samples that were inspected blindly, variety of illuminating genes, and computational time. Support vector machines and other classifiers have been used to evaluate the usefulness and define the characteristics of the identified genes [79].
Ansari et al. offered a technique for identifying cancer by using image analysis and PSO SVM. The mathematician’s elimination filter is used during the picture preparation step to eliminate noise from images. Identifying objects in a photo and, most importantly, the regions of interest is made easier with the help of proposed image segmentation. To extract relevant data from digital photos, the PCA method is used. Following that, PSO with SVM as a wrapper method was used for classification. Higher levels of precision, sensitivity, and specificity were founded in this proposed PSO with SVM wrapper approach [80].
In order to choose the most precise and useful genes for cancer categorization, Houssein et al. presented the barnacle mating optimizer technique enhanced with SVM for microarray gene expression profiling. Two microarray datasets were used to evaluate the effectiveness of the proposed model: a binary microarray dataset (i.e., leukemia1) and a multi-class microarray dataset (i.e., SRBCT, lymphoma, and leukemia2). The Tunicate Swarm Algorithm, Genetic Algorithm, Particle Swarm Optimization, and Artificial Bee Colony were only a few well-known meta-heuristic optimization strategies that were defeated in the testing by the suggested BMO-SVM methodology. It is crucial to note that our suggested method outperforms existing methods in terms of information superiority [81].

6. Discussion

This review is based on 97 articles collected and studied on dimension reduction for gene selection problems in cancer classification and prediction. The developments in gene expression technology have contributed to faster research on multiple genes at the same time. High-dimensional gene expression data, on the other hand, do not appear to be capable of producing a significant and relevant number of predictable gene expressions. In this review, we analyzed several different papers to find out about various dimension reduction methods to find the best set of genes that help to classify cancer disease. There are only a few techniques that are able to handle high-dimensional gene expression data, but a large percentage of papers for the same have been published (Figure 9). The review also shows that nature-inspired algorithms are frequently used to reduce the dimension of gene expression data. Some of the novel hybrid techniques that were developed through the combination of nature-inspired algorithms provided enough improvement in cancer classification and prediction [82,83,84,85,86,87].
As the review shows, nature-inspired approaches achieved big contributions and clear success in solving the problems of gene selection, cancer classification, and prediction. It was clear from the analysis that the hybridization modification technique with a nature-inspired algorithm was applied in 67 references to solve the problem of gene selection. This counting result shows that hybridization is the most widely applied modification technique to enhance nature-inspired algorithms and reduce the cost of the wrapper approach in the gene selection domain for cancer classification and prediction. The high number of researcher work comes from GA, which had the greatest number of works regarding wrapper hybridization. PSO was also hybridized in most of the references; conversely ACO’s hybridization works were fewer. The PSO and ABC are also suitable for dimension reduction for further analysis with hybridization. It was also noticed that hybrid techniques with different combinations were the most prominent techniques for the classifiers [88,89,90,91,92]. Figure 10 shows the number of publications and citations for each nature-inspired algorithm among the main publishers.

Implementation Challenges of a Nature-Inspired Algorithm for Gene Expression Data for Cancer Classification and Prediction

The nature-inspired framework enables gene emergency monitoring and diagnosis of cancer through big data of gene expression with ML technologies. There is a growing expectation that ML and deep learning models will help improve diagnostic procedures with nature-inspired algorithms. However, there are many implementation challenges associated with the collection and processing of massive amounts of cancer gene expression big data to understand patients’ problems and then diagnose cancer through sophisticated AI and ML algorithms. In this section, discuss the implementation challenges of the nature-inspired algorithms for gene expression data analysis. Some of the major implementation challenges are represented below (Figure 11).
Data Pre-processing.
Data preparation is the most essential step in the development of effective ML models with a nature-inspired algorithm for cancer detection challenges. Data gathered from various sources is in a variety of formats, including structured, semi-structured, and unstructured, and cannot be utilized directly in ML models. Therefore, various pre-processing techniques are required before applying the nature-inspired algorithm for further procedures. Pre-processing techniques transform data into a more precise format that can be easily implemented with nature-inspired frameworks for accurate cancer detection and prediction.
Imbalance Data.
Class imbalance occurs when a dataset contains disproportionately more records for one class than the other. Most machine learning models are biased toward the majority class. As a result, models frequently tend to focus on accuracy rather than identifying rare events, such as a patient’s emergency state. To handle class imbalance, different methods can be used, such as kernel and cost-based methods, and sampling. One of the challenges with a nature-inspired framework is how to choose the best method to face the class imbalance problem of bio-medical data for cancer detection and prediction.
Selection of Classifiers.
The computational effectiveness of gene selection and the accuracy of gene subset predictions are both considered by hybrid approaches. Two stages comprise a hybrid method’s general framework. First, to reduce the size of a given gene collection without losing possible discriminating information, less significant genes are eliminated using an independent-classifier criterion. Following that, various classifiers are used to identify relevant genes for prediction performance using a variety of evaluation metrics. Traditional selection techniques mostly concentrate on a single gene subset with one classifier, which makes a small contribution to the task of prediction. As a result, numerous gene selection methods have been developed to pick important genes using different classifiers. Therefore, the most difficult task is to select the appropriate classifiers for all the nature-inspired frameworks of gene selection.
Learning Process and Size of Training Data.
Medical data collected from various sources is high-dimensional and has an unequal distribution of classes. The best cross-validation techniques are also necessary to enhance the performance of ML and deep learning algorithms. The learning process with various sizes of training data is one of the most important necessities in solving cancer detection challenges. Due to several computational and statistical challenges of nature-inspired algorithms, the learning process of the framework is a more challenging task with high-dimensional cancer gene expression data.
Performance of the Exploration phase and the Exploitation Phase.
For the successful implementation of nature-inspired algorithms, the balance between the exploration phase and exploitation phase is an important step. Some nature-inspired algorithms are good in the exploration phase, while others are good in the exploitation phaseto optimize the performance of a hybrid, nature-inspired framework for classifying and predicting cancer. An important and challenging step is finding the right balance between the exploration and exploitation phases.

7. Advantages and Disadvantages of Nature-Inspired Algorithms

This review presented a survey on dimension reduction using nature-inspired algorithms to address gene selection for cancer classification and prediction. The review is based on a solid theoretical, applied, and technical foundation for the dimension reduction of gene expression biomedical data. Three main research streams are identified in this review: dimension reduction techniques for high-dimensional gene expression data, a number of selected nature-inspired algorithms are systematically reviewed for the same, followed by the original principle behind each of the nature-inspired algorithms. This review aims to draw the map for researchers and guide them when creating new research in this area. Hybridization is the most widely used modification technique for the dimension reduction problem. There are three types of hybridization: integrating a nature-inspired algorithm with another nature-inspired algorithm, integrating a nature-inspired algorithm with a classifier, and integrating a nature-inspired algorithm with filter or extraction techniques [92,93,94]. Some advantages and disadvantages of nature-inspired algorithms are given below.
Advantages:
  • These algorithms are very effective in locating multi-dimensional and multi-modal issues in high-dimensional gene expression data and finding the optimal solutions.
  • Regarding applications of nature-inspired algorithms for gene selection, it was evident that microarray gene expression classification is the most dominant application where nature-inspired algorithms for gene selection are applied and successfully resolve other applications of biomedical data.
  • These algorithms have been employed successfully to find different human disorders.
  • Compared to the prior option, nature-inspired algorithms make it easier to recognize rational-universal issues in an adequate time frame with greater reliability and precision.
Disadvantages:
Despite the efficiency of meta-heuristics in tackling challenging optimization problems, some obstacles impact their performance. These include dynamicity, multi-objectivity, constraint, the problem of search space, and uncertainty. The performance of the algorithm under discussion can be impacted by the analysis tool parameter value that is included in all nature-inspired algorithms. The optimal settings or numbers to use or how to adjust these parameters to get the greatest performance remain unclear [94,95,96,97].
  • In the case of high-dimensional biomedical data, the real environments are complicated, and the optimization problems can be high-dimensional, large-scale, multi-modal, and multi-objective; the optimization environments can be dynamic, highly constrained, and uncertain; the fitness evaluations may contain noise and be imprecise and time-consuming.
  • The complexity of real environments poses a great challenge to nature-inspired algorithms. Although some researchers have made attempts to solve the aforementioned problems, figuring out how to handle these issues remains a very difficult problem.
  • Nature-inspired algorithms currently need a complete mathematical framework for analyzing all methods to fully understand their robustness, consistency, development, and levels of integration.
  • Due to the probabilistic character of nature-inspired methods, results are not entirely repeatable; hence, several runs are necessary to acquire useful data.

Take-Home Message of the Review

The accurate prediction of the presence of a specific form of cancer is greatly assisted by the quick development of feature selection algorithms with nature-inspired machine-learning approaches in the field of medical diagnosis. The creation of various hybrid models with nature-inspired algorithms for cancer classification would be beneficial for the reduction of obstacles to further research as well as an improvement in classification performance and computing efficiency. Figure 12 shows the overall hybridization process of a nature-inspired algorithm to optimize the performance of ML for cancer classification and prediction or to develop the new rule mining for further analysis on high-dimensional biomedical data.

8. Conclusions

This paper has presented different ways of reducing the dimensionality of high-dimensional gene expression data. The increase in the amount of data to be analyzed has made dimensionality reduction methods essential in order to get meaningful results. Different feature selection and feature extraction methods were described and compared. Their advantages and disadvantages were also discussed. In addition, we presented several of the most popular nature-inspired algorithms for biomedical data and their review for cancer classification and prediction. This survey is an effort to provide a research repository and a useful reference for researchers to guide them when planning to develop a new hybrid framework with nature-inspired algorithms to solve gene selection problems for further analysis. Through this review paper, we identified the key issues that researchers faced for gene selection from high-dimensional biomedical data with nature-inspired algorithms and also examined modifications that are required to tackle the gene selection problem with nature-inspired algorithms. In this work, we also observed the fundamental idea and reasoning behind gene selection for cancer classification and prediction from high-dimensional gene expression data. Outstanding challenges of dealing with high-dimensional gene expression datasets and future works are listed and briefly explained, which may help researchers draw a roadmap on recent trends and potential future research directions. Future Directions.
This review demonstrates how vast and rapidly expanding the field of computing is influenced by using nature-inspired algorithms. With emphasis, this invited paper focuses on biologically based algorithms and methodologies and offers a short assessment of key advancements built in this interesting field of study. Practitioners may benefit from having a basic understanding of these algorithms, especially those in the information or data science industries. They might use this as a jumping-off point for further reading to investigate many of these bio-inspired computing algorithms for practical applications in social, organizational, or governance problems, depending on the particulars of their business problem and the complications associated with the objectives, constraints, and computational space. Biomedical microarray gene expression datasets offer too many dimensions for the small number of samples. Our research brings us to the conclusion that data dimensionality has a massive effect on a diagnostic system’s accuracy in many circumstances. In order to deal with large datasets, effective algorithms are required that minimize redundancy and dependency while preserving informative genes. This work is intended to perform a comprehensive comparative analysis of well-known nature-inspired algorithms to reduce the dimensionality of biomedical data. A hybrid gene selection method with nature-inspired algorithms is a new trend in cancer classification and prediction techniques based on high-dimensional gene expression data. The most widely used framework is the one that integrates a classifier with the nature-inspired algorithm in the wrapper phase. Microarray and medical applications are the dominant applications where most of the nature-inspired algorithms are modified and used for gene selection. Despite the popularity of nature-inspired algorithms for gene selection, there are still many areas that need further investigation. Furthermore, it can be imagined that a thorough investigation and improvement of nature-inspired algorithms will improve the feature selection process in various high-dimensional areas. This review paper will be used to help researchers develop nature-inspired algorithms for tackling the dimension reduction problem for cancer classification and prediction.

Author Contributions

Conceptualization, A.Y. and R.M.A.; methodology, A.Y., R.M.A. and P.L.; formal analysis, A.Y., R.M.A., P.L., A.M. and N.K.V.; resources, A.Y. and R.M.A.; validation, P.L., N.K.V. and A.M.; writing, original draft preparation, A.Y. and R.M.A.; writing review and editing, A.Y., R.M.A., P.K. and N.K.V.; supervision, A.Y., R.M.A. and P.K. All authors have read and agreed to the published version of the manuscript.

Funding

No funding was received for this research.

Data Availability Statement

Not applicable.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Shah, S.M.; Khan, R.A.; Arif, S.; Sajid, U. Artificial intelligence for breast cancer analysis: Trends & directions. Comput. Biol. Med. 2022, 15, 105221. [Google Scholar] [CrossRef]
  2. Aziz, R.; Verma, C.K.; Jha, M.; Srivastava, N. Artificial neural network classification of microarray data using new hybrid gene selection method. Int. J. Data Min. Bioinform. 2017, 17, 42–65. [Google Scholar] [CrossRef]
  3. Tufail, A.B.; Ma, Y.K.; Kaabar, M.K.; Martínez, F.; Junejo, A.R.; Ullah, I.; Khan, R. Deep learning in cancer diagnosis and prognosis prediction: A minireview on challenges, recent trends, and future directions. Comput. Math. Methods Med. 2021, 2021, 9025470. [Google Scholar] [CrossRef] [PubMed]
  4. Diao, R.; Shen, Q. Nature inspired feature selection meta-heuristics. Artif. Intell. Rev. 2015, 44, 311–340. [Google Scholar] [CrossRef] [Green Version]
  5. Aziz, R.M. Cuckoo search-based optimization for cancer classification: A new hybrid approach. J. Comput. Biol. 2022, 29, 565–584. [Google Scholar] [CrossRef]
  6. Elemento, O.; Leslie, C.; Lundin, J.; Tourassi, G. Artificial intelligence in cancer research, diagnosis and therapy. Nat. Rev. Cancer 2021, 21, 747–752. [Google Scholar] [CrossRef]
  7. Aziz, R.; Verma, C.K.; Srivastava, N. Dimension reduction methods for microarray data: A review. AIMS Bioeng. 2017, 4, 179–197. [Google Scholar] [CrossRef]
  8. Alomari, O.A.; Khader, A.T.; Al-Betar, M.A.; Abualigah, L.M. Gene selection for cancer classification by combining minimum redundancy maximum relevancy and bat-inspired algorithm. Int. J. Data Min. Bioinform. 2017, 19, 32–51. [Google Scholar] [CrossRef]
  9. Aziz, R.M.; Mahto, R.; Goel, K.; Das, A.; Kumar, P.; Saxena, A. Modified Genetic Algorithm with Deep Learning for Fraud Transactions of Ethereum Smart Contract. Appl. Sci. 2023, 13, 697. [Google Scholar] [CrossRef]
  10. Musheer, R.A.; Verma, C.K.; Srivastava, N. Novel machine learning approach for classification of high-dimensional microarray data. Soft Comput. 2019, 23, 13409–13421. [Google Scholar] [CrossRef]
  11. Ramachandran, A.; Rustum, R.; Adeloye, A.J. Review of anaerobic digestion modeling and optimization using nature-inspired techniques. Processes 2019, 12, 953. [Google Scholar] [CrossRef] [Green Version]
  12. Mohamad, A.B.; Zain, A.M.; Nazira Bazin, N.E. Cuckoo search algorithm for optimization problems a literature review and its applications. Appl. Artif. Intell. 2014, 28, 419–448. [Google Scholar] [CrossRef]
  13. Aziz, R.; Verma, C.K.; Srivastava, N. Artificial neural network classification of high dimensional data with novel optimization approach of dimension reduction. Ann. Data Sci. 2018, 5, 615–635. [Google Scholar] [CrossRef]
  14. Naik, M.; Nath, M.R.; Wunnava, A.; Sahany, S.; Panda, R. A new adaptive Cuckoo search algorithm. In Proceedings of the 2015 IEEE 2nd International Conference on Recent Trends in Information Systems (ReTIS), Kolkata, India, 9–11 July 2015; IEEE: New York, NY, USA; pp. 1–5. [Google Scholar] [CrossRef]
  15. Heyn, P.C.; Meeks, S.; Pruchno, R. Methodological guidance for a quality review article. Gerontologist 2019, 59, 197–201. [Google Scholar] [CrossRef] [Green Version]
  16. Yildiz, A.R. Cuckoo search algorithm for the selection of optimal machining parameters in milling operations. Int. J. Adv. Manuf. Technol. 2013, 64, 55–61. [Google Scholar] [CrossRef]
  17. Fister, I.; Yang, X.S.; Fister, D.; Fister, I. Cuckoo search: A brief literature review. In Cuckoo Search and Firefly Algorithm: Theory and Applications; Springer: Cham, Switzerland, 2014; pp. 49–62. [Google Scholar] [CrossRef] [Green Version]
  18. Civicioglu, P.; Besdok, E. Comparative analysis of the cuckoo search algorithm. In Cuckoo Search and Firefly Algorithm: Theory and Applications; Springer: Cham, Switzerland, 2014; pp. 85–113. [Google Scholar] [CrossRef]
  19. Aziz, R.M.; Baluch, M.F.; Patel, S.; Ganie, A.H. LGBM: A approach for Ethereum fraud detection. Int. J. Inf. Technol. 2022, 14, 3321–3331. [Google Scholar] [CrossRef]
  20. Aziz, R.; Verma, C.K.; Srivastava, N. A novel approach for dimension reduction of microarray. Comput. Biol. Chem. 2017, 71, 161–169. [Google Scholar] [CrossRef]
  21. Aziz, R.; Srivastava, N.; Verma, C.K. T-independent component analysis for svm classification of DNA-microarray data. Int. J. Bioinform. Res. 2015, 6, 305–312. [Google Scholar] [CrossRef]
  22. Desai, N.P.; Baluch, M.F.; Makrariya, A.; Musheer Aziz, R. Image processing model with deep learning approach for fish species classification. Turk. J. Comput. Math. Educ. 2022, 13, 85–99. [Google Scholar]
  23. Aziz, R.M.; Hussain, A.; Sharma, P.; Kumar, P. Machine learning based soft computing regression analysis approach for crime data prediction. Karbala Int. J. Mod. Sci. 2022, 8, 1–19. [Google Scholar] [CrossRef]
  24. Aziz, R.M.; Baluch, M.F.; Patel, S.; Kumar, P. A based approach to detect the Ethereum fraud transactions with limited attributes. Karbala Int. J. Mod. Sci. 2022, 8, 139–151. [Google Scholar] [CrossRef]
  25. Luan, J.; Yao, Z.; Zhao, F.; Song, X. A novel method to solve supplier selection problem: Hybrid algorithm of genetic algorithm and ant colony optimization. Math. Comput. Simul. 2019, 156, 294–309. [Google Scholar] [CrossRef]
  26. Aziz, R.M.; Sharma, P.; Hussain, A. Algorithms for Crime Prediction under Indian Penal Code. Ann. Data Sci. 2022, 1–32. [Google Scholar] [CrossRef]
  27. Deepak, M.; Rustum, R. Review of Latest Advances in Nature-Inspired Algorithms for Optimization of Activated Sludge Processes. Processes 2022, 11, 77. [Google Scholar] [CrossRef]
  28. Sagu, A.; Gill, N.S.; Gulia, P.; Singh, P.K.; Hong, W.C. Design of Metaheuristic Optimization Algorithms for Deep Learning Model for Secure IoT Environment. Sustainability 2023, 15, 2204. [Google Scholar] [CrossRef]
  29. Aziz, R.; Verma, C.; Srivastava, N. A weighted-SNR feature selection from independent component subspace for nb classification of microarray data. Int. J. Biotechnol. Res. 2015, 6, 245–255. [Google Scholar] [CrossRef]
  30. Prabukumar, M.; Agilandeeswari, L.; Ganesan, K. An intelligent lung cancer diagnosis system using cuckoo search optimization and support vector machine classifier. J. Ambient Intell. Humaniz. Comput. 2019, 10, 267–293. [Google Scholar] [CrossRef]
  31. Alzaqebah, M.; Briki, K.; Alrefai, N.; Brini, S.; Jawarneh, S.; Alsmadi, M.K.; Mohammad, R.M.A.; ALmarashdeh, I.; Alghamdi, F.A.; Aldhafferi, N.; et al. Memory based cuckoo search algorithm for feature selection of gene expression dataset. Inform. Med. Unlocked 2021, 24, 100572. [Google Scholar] [CrossRef]
  32. Gunavathi, C.; Premalatha, K. Cuckoo search optimisation for feature selection in cancer classification: A new approach. Int. J. Data Min. Bioinform. 2015, 13, 248–265. [Google Scholar] [CrossRef]
  33. Arjmand, A.; Meshgini, S.; Afrouzian, R.; Farzamnia, A. Breast tumor segmentation using K-means clustering and cuckoo search optimization. In Proceedings of the 2019 9th International Conference on Computer and Knowledge Engineering (ICCKE), Mashhad, Iran, 24–25 October 2019; IEEE: New York, NY, USA; pp. 305–308. [Google Scholar] [CrossRef]
  34. Sampathkumar, A.; Rastogi, R.; Arukonda, S.; Shankar, A.; Kautish, S.; Sivaram, M. An efficient hybrid methodology for detection of cancer-causing gene using CSC for micro array data. J. Ambient Intell. Humaniz. Comput. 2020, 11, 4743–4751. [Google Scholar] [CrossRef]
  35. Akila, S.; Christe, S.A. A wrapper based binary bat algorithm with greedy crossover for attribute selection. Expert Syst. Appl. 2022, 187, 115828. [Google Scholar] [CrossRef]
  36. Al-Betar, M.A.; Alomari, O.A.; Abu-Romman, S.M. A TRIZ-inspired bat algorithm for gene selection in cancer classification. Genomics 2020, 112, 114–126. [Google Scholar] [CrossRef]
  37. Jeyasingh, S.; Veluchamy, M. Modified bat algorithm for feature selection with the wisconsin diagnosis breast cancer (WDBC) dataset. Asian Pac. J. Cancer Prev. 2017, 18, 1257. [Google Scholar] [CrossRef]
  38. Alhassan, A.M.; Zainon, W.M.N.W. BAT algorithm with fuzzy C-ordered means (BAFCOM) clustering segmentation and enhanced capsule networks (ECN) for brain cancer MRI images classification. IEEE Access 2020, 8, 201741–201751. [Google Scholar] [CrossRef]
  39. Hambali, M.A.; Oladele, T.O.; Adewole, K.S.; Sangaiah, A.K.; Gao, W. Feature selection and computational optimization in high-dimensional microarray cancer datasets via InfoGain-modified bat algorithm. Multimed. Tools Appl. 2022, 81, 36505–36549. [Google Scholar] [CrossRef]
  40. Chatra, K.; Kuppili, V.; Edla, D.R.; Verma, A.K. Cancer data classification using binary bat optimization and extreme learning machine with a novel fitness function. Med. Biol. Eng. Comput. 2019, 57, 2673–2682. [Google Scholar] [CrossRef]
  41. Ahmad, F.; Mat Isa, N.A.; Hussain, Z.; Osman, M.K.; Sulaiman, S.N. A GA-based feature selection and parameter optimization of an ANN in diagnosing breast cancer. Pattern Anal. Appl. 2015, 18, 861–870. [Google Scholar] [CrossRef]
  42. Mohammadi, S.; Hejazi, S.R. Using particle swarm optimization and genetic algorithms for optimal control of non-linear fractional-order chaotic system of cancer cells. Math. Comput. Simul. 2023, 206, 538–560. [Google Scholar] [CrossRef]
  43. Rojas, M.G.; Olivera, A.C.; Carballido, J.A.; Vidal, P.J. Memetic micro-genetic algorithms for cancer data classification. Intell. Syst. Appl. 2023, 17, 200173. [Google Scholar] [CrossRef]
  44. Zexuan, Z.; Ong, Y.S.; Dash, M. Markov blanket-embedded genetic algorithm for gene selection. Pattern Recognit. 2007, 40, 3236–3248. [Google Scholar] [CrossRef]
  45. Motieghader, H.; Najafi, A.; Sadeghi, B.; Masoudi-Nejad, A. A hybrid gene selection algorithm for microarray cancer classification using genetic algorithm and learning automata. Inform. Med. Unlocked 2017, 9, 246–254. [Google Scholar] [CrossRef]
  46. Jansi Rani, M.; Devaraj, D. Two-stage hybrid gene selection using mutual information and genetic algorithm for cancer data classification. J. Med. Syst. 2019, 43, 235. [Google Scholar] [CrossRef] [PubMed]
  47. Paul, D.; Su, R.; Romain, M.; Sébastien, V.; Pierre, V.; Isabelle, G. Feature selection for outcome prediction in oesophageal cancer using genetic algorithm and random forest classifier. Comput. Med. Imaging Graph. 2017, 60, 42–49. [Google Scholar] [CrossRef] [PubMed]
  48. Lu, C.; Zhu, Z.; Gu, X. An intelligent system for lung cancer diagnosis using a new genetic algorithm based feature selection method. J. Med. Syst. 2018, 38, 97. [Google Scholar] [CrossRef] [PubMed]
  49. Abbas, S.; Jalil, Z.; Javed, A.R.; Batool, I.; Khan, M.Z.; Noorwali, A.; Gadekallu, T.R.; Akbar, A. BCD-WERT: A novel approach for breast cancer detection using whale optimization based efficient features and extremely randomized tree algorithm. PeerJ Comput. Sci. 2021, 7, e390. [Google Scholar] [CrossRef]
  50. Fang, H.; Fan, H.; Lin, S.; Qing, Z.; Sheykhahmad, F.R. Automatic breast cancer detection based on optimized neural network using whale optimization algorithm. Int. J. Imaging Syst. Technol. 2021, 31, 425–438. [Google Scholar] [CrossRef]
  51. Stephan, P.; Stephan, T.; Kannan, R.; Abraham, A. A hybrid artificial bee colony with whale optimization algorithm for improved breast cancer diagnosis. Neural Comput. Appl. 2021, 33, 13667–13691. [Google Scholar] [CrossRef]
  52. Sayed, G.I.; Darwish, A.; Hassanien, A.E. Binary whale optimization algorithm and binary moth flame optimization with clustering algorithms for clinical breast cancer diagnoses. J. Classif. 2020, 37, 66–96. [Google Scholar] [CrossRef]
  53. Sayed, G.I.; Darwish, A.; Hassanien, A.E.; Pan, J.S. Breast cancer diagnosis approach based on meta-heuristic optimization algorithm inspired by the bubble-net hunting strategy of whales. In Genetic and Evolutionary Computing, Proceedings of the Tenth International Conference on Genetic and Evolutionary Computing, Fuzhou, China, 7–9 November 2016; Springer International Publishing: Cham, Switzerland, 2017; pp. 306–313. [Google Scholar] [CrossRef]
  54. Kaur, N.; Kaur, L.; Cheema, S.S. An enhanced version of Harris Hawks optimization by dimension learning-based hunting for breast cancer detection. Sci. Rep. 2021, 11, 21933. [Google Scholar] [CrossRef]
  55. Rammurthy, D.; Mahesh, P.K. Whale Harris hawks optimization based deep learning classifier for brain tumor detection using MRI images. J. King Saud Univ. Comput. Inf. Sci. 2022, 34, 3259–3272. [Google Scholar] [CrossRef]
  56. Ibrahim, H.T.; Mazher, W.J.; Jassim, E.M. Modified Harris Hawks optimizer for feature selection and support vector machine kernels. Indones. J. Electr. Eng. Comput. Sci. 2023, 29, 942–953. [Google Scholar]
  57. Hussien, A.G.; Abualigah, L.; Abu Zitar, R.; Hashim, F.A.; Amin, M.; Saber, A.; Almotairi, K.H.; Gandomi, A.H. Recent advances in harris hawks optimization: A comparative study and applications. Electronics 2022, 11, 1919. [Google Scholar] [CrossRef]
  58. Jiang, F.; Zhu, Q.; Tian, T. Breast Cancer Detection Based on Modified Harris Hawks Optimization and Extreme Learning Machine Embedded with Feature Weighting. Neural Process. Lett. 2022, 23, 1–24. [Google Scholar] [CrossRef]
  59. Sun, L.; Kong, X.; Xu, J.; Xue, Z.A.; Zhai, R.; Zhang, S. A hybrid gene selection method based on ReliefF and ant colony optimization algorithm for tumor classification. Sci. Rep. 2019, 9, 8978. [Google Scholar] [CrossRef] [Green Version]
  60. Sumeyya, İ.; Gençtürk, T.H.; Gülağız, F.K.; Özcan, H.; Altuncu, M.A.; Şahin, S. hybSVM: Bacterial colony optimization algorithm based SVM for malignant melanoma detection. Eng. Sci. Technol. Int. J. 2021, 24, 1059–1071. [Google Scholar] [CrossRef]
  61. Fallahzadeh, O.; Dehghani-Bidgoli, Z.; Assarian, M. Raman spectral feature selection using ant colony optimization for breast cancer diagnosis. Lasers Med. Sci. 2018, 33, 1799–1806. [Google Scholar] [CrossRef]
  62. Zainuddin, S.; Nhita, F.; Wisesty, U.N. Classification of gene expressions of lung cancer and colon tumor using Adaptive-Network-Based Fuzzy Inference System (ANFIS) with Ant Colony Optimization (ACO) as the feature selection. J. Phys. Conf. Ser. 2019, 1192, 012019. [Google Scholar] [CrossRef]
  63. Rajagopal, R. Glioma brain tumor detection and segmentation using weighting random forest classifier with optimized ant colony features. Int. J. Imaging Syst. Technol. 2019, 29, 353–359. [Google Scholar] [CrossRef]
  64. Coleto-Alcudia, V.; Vega-Rodríguez, M.A. Artificial bee colony algorithm based on dominance (ABCD) for a hybrid gene selection method. Knowl. Based Syst. 2020, 205, 106323. [Google Scholar] [CrossRef]
  65. Aziz, R.M. Application of nature inspired soft computing techniques for gene selection: A novel frame work for classification of cancer. Soft Comput. 2022, 26, 12179–12196. [Google Scholar] [CrossRef]
  66. Punitha, S.; Al-Turjman, F.; Stephan, T. An automated breast cancer diagnosis using feature selection and parameter optimization in ANN. Comput. Electr. Eng. 2021, 90, 106958. [Google Scholar] [CrossRef]
  67. Karthiga, B.; Rekha, M. Feature extraction and I-NB classification of CT images for early lung cancer detection. Mater. Today Proc. 2020, 33, 3334–3341. [Google Scholar] [CrossRef]
  68. Aljanabi, M.; Jumaa, F.A.; Abed, J.K.; Al-Hamadani, H. Analysis of Automatic Detection of Tumour Lesions Images using Bee Colony Technique. J. Phys. Conf. Ser. 2020, 1530, 012012. [Google Scholar] [CrossRef]
  69. Al-Thanoon, N.A.; Qasim, O.S.; Algamal, Z.Y. Tuning parameter estimation in SCAD-support vector machine using firefly algorithm with application in gene selection and cancer classification. Comput. Biol. Med. 2018, 103, 262–268. [Google Scholar] [CrossRef] [PubMed]
  70. Sawhney, R.; Mathur, P.; Shankar, R. A firefly algorithm based wrapper-penalty feature selection method for cancer diagnosis. In Computational Science and Its Applications–ICCSA 2018, Proceedings of the 18th International Conference, Melbourne, VIC, Australia, 2–5 July 2018; Proceedings, Part I 18; Springer International Publishing: Cham, Switzerland, 2018; pp. 438–449. [Google Scholar] [CrossRef]
  71. Farouk, R.M.; Mustafa, H.I.; Ali, A.E. Hybrid Firefly and Swarm Algorithms for Breast Cancer Mammograms Classification Based on Rough Set Theory Features Selection. In Proceedings of the Future Technologies Conference (FTC), Vancouver, BC, Canada, 28–29 October 2021; Springer International Publishing: Cham, Switzerland; Volume 3, pp. 849–867. [Google Scholar] [CrossRef]
  72. Alshamlan, H.M.; Badr, G.H.; Alohali, Y.A. Genetic Bee Colony (GBC) algorithm: A new gene selection method for microarray cancer classification. Comput. Biol. Chem. 2015, 56, 49–60. [Google Scholar] [CrossRef]
  73. Nguyun, L.C.; Meesad, P.; Unger, H. A highly accurate firefly based algorithm for heart disease prediction. Expert Syst. Appl. 2015, 42, 8221–8231. [Google Scholar] [CrossRef]
  74. Banu, P.N.; Azar, A.T.; Inbarani, H.H. Fuzzy firefly clustering for tumour and cancer analysis. Int. J. Model. Identif. Control. 2017, 27, 92–103. [Google Scholar] [CrossRef]
  75. Almugren, N.; Alshamlan, H.M. New bio-marker gene discovery algorithms for cancer gene expression profile. IEEE Access 2019, 7, 136907–136913. [Google Scholar] [CrossRef]
  76. Osama, S.; Shaban, H.; Ali, A.A. Gene reduction and algorithms for cancer classification based on microarray gene expression data: A comprehensive review. Expert Syst. Appl. 2022, 213, 118946. [Google Scholar] [CrossRef]
  77. Hu, Y.; Zhang, Y.; Gao, X.; Gong, D.; Song, X.; Guo, Y.; Wang, J. A federated feature selection algorithm based on particle swarm optimization under privacy protection. Knowl. Based Syst. 2023, 260, 110122. [Google Scholar] [CrossRef]
  78. Al-Thanoon, N.A.; Qasim, O.S.; Algamal, Z.Y. A new hybrid firefly algorithm and particle swarm optimization for tuning parameter estimation in penalized support vector machine with application in chemometrics. Chemom. Intell. Lab. Syst. 2019, 184, 142–152. [Google Scholar] [CrossRef]
  79. Kar, S.; Sharma, K.D.; Maitra, M. Gene selection from microarray gene expression data for classification of cancer subgroups employing PSO and adaptive K-nearest neighborhood technique. Expert Syst. Appl. 2015, 42, 612–627. [Google Scholar] [CrossRef]
  80. Ansari, A.S.; Zamani, A.S.; Mohammadi, M.S.; Ritonga, M.; Ahmed, S.S.; Pounraj, D.; Kaliyaperumal, K. Detection of Pancreatic Cancer in CT Scan Images Using PSO SVM and Image Processing. BioMed Res. Int. 2022, 2022, 8544337. [Google Scholar] [CrossRef]
  81. Houssein, E.H.; Abdelminaam, D.S.; Hassan, H.N.; Al-Sayed, M.M.; Nabil, E. A hybrid barnacles mating optimizer algorithm with support vector machines for gene selection of microarray cancer classification. IEEE Access 2021, 9, 64895–64905. [Google Scholar] [CrossRef]
  82. Kaul, S.; Kumar, Y.; Ghosh, U.; Alnumay, W. Nature-inspired optimization algorithms for different computing systems: Novel perspective and systematic review. Multimed. Tools Appl. 2022, 81, 26779–26801. [Google Scholar] [CrossRef]
  83. Lazebnik, T. Cell-Level Spatio-Temporal Model for a Bacillus Calmette–Guérin-Based Immunotherapy Treatment Protocol of Superficial Bladder Cancer. Cells 2022, 11, 2372. [Google Scholar] [CrossRef]
  84. Vahmiyan, M.; Kheirabadi, M.; Akbari, E. Feature selection methods in microarray gene expression data: A systematic mapping study. Neural Comput. Appl. 2022, 34, 19675–19702. [Google Scholar] [CrossRef]
  85. Abu Khurma, R.; Aljarah, I.; Sharieh, A.; Abd Elaziz, M.; Damaševičius, R.; Krilavičius, T. A review of the modification strategies of the nature inspired algorithms for feature selection problem. Mathematics 2022, 10, 464. [Google Scholar] [CrossRef]
  86. Nave, O.; Elbaz, M. Artificial immune system features added to breast cancer clinical data for machine learning (ML) applications. Biosystems 2021, 202, 104341. [Google Scholar] [CrossRef]
  87. Almazrua, H.; Alshamlan, H. A comprehensive survey of recent hybrid feature selection methods in cancer microarray gene expression data. IEEE Access 2022, 10, 71427–71449. [Google Scholar] [CrossRef]
  88. Alrefai, N.; Ibrahim, O. Optimized feature selection method using particle swarm intelligence with ensemble learning for cancer classification based on microarray datasets. Neural Comput. Appl. 2022, 34, 13513–13528. [Google Scholar] [CrossRef]
  89. Nassif, A.B.; Talib, M.A.; Nasir, Q.; Afadar, Y.; Elgendy, O. Breast cancer detection using artificial intelligence techniques: A systematic literature review. Artif. Intell. Med. 2022, 127, 102276. [Google Scholar] [CrossRef] [PubMed]
  90. Lazebnik, T.; Bunimovich-Mendrazitsky, S. Improved Geometric Configuration for the Bladder Cancer BCG-Based Immunotherapy Treatment Model. In Mathematical and Computational Oncology, Proceedings of the Third International Symposium, ISMCO 2021, Virtual Event, 11–13 October 2021; Springer International Publishing: Cham, Switzerland, 2021; pp. 63–67. [Google Scholar] [CrossRef]
  91. Aziz, R.M.; Desai, N.P.; Baluch, M.F. Computer vision model with novel cuckoo search based deep learning approach for classification of fish image. Multimed. Tools Appl. 2022, 82, 3677–3696. [Google Scholar] [CrossRef]
  92. Adhikari, M.; Srirama, S.N.; Amgoth, T. A comprehensive survey on nature-inspired algorithms and their applications in edge computing: Challenges and future directions. Softw. Pract. Exp. 2022, 52, 1004–1034. [Google Scholar] [CrossRef]
  93. Guzev, E.; Halachmi, S.; Bunimovich-Mendrazitsky, S. Additional extension of the mathematical model for BCG immunotherapy of bladder cancer and its validation by auxiliary tool. Int. J. Nonlinear Sci. Numer. Simul. 2019, 20, 675–689. [Google Scholar] [CrossRef]
  94. Aziz, R.M. Nature-inspired metaheuristics model for gene selection and classification of biomedical microarray data. Med. Biol. Eng. Comput. 2022, 60, 1627–1646. [Google Scholar] [CrossRef]
  95. Hameed, S.S.; Hassan, W.H.; Latiff, L.A.; Muhammadsharif, F.F. A comparative study of nature-inspired metaheuristic algorithms using a three-phase hybrid approach for gene selection and classification in high-dimensional cancer datasets. Soft Comput. 2021, 25, 8683–8701. [Google Scholar] [CrossRef]
  96. Ghosh, M.; Sen, S.; Sarkar, R.; Maulik, U. Quantum squirrel inspired algorithm for gene selection in methylation and expression data of prostate cancer. Appl. Soft Comput. 2021, 105, 107221. [Google Scholar] [CrossRef]
  97. Jain, I.; Jain, V.K.; Jain, R. Correlation feature selection based improved-binary particle swarm optimization for gene selection and cancer classification. Appl. Soft Comput. 2018, 62, 203–215. [Google Scholar] [CrossRef]
Figure 1. The PRISMA systematic flow of selection strategy for the study.
Figure 1. The PRISMA systematic flow of selection strategy for the study.
Mathematics 11 01081 g001
Figure 2. Taxonomy of dimension reduction algorithms for biomedical data analysis.
Figure 2. Taxonomy of dimension reduction algorithms for biomedical data analysis.
Mathematics 11 01081 g002
Figure 3. Benefits of the dimension reduction method.
Figure 3. Benefits of the dimension reduction method.
Mathematics 11 01081 g003
Figure 4. Widely used nature-inspired algorithms.
Figure 4. Widely used nature-inspired algorithms.
Mathematics 11 01081 g004
Figure 5. The working procedure of nature-inspired algorithms.
Figure 5. The working procedure of nature-inspired algorithms.
Mathematics 11 01081 g005
Figure 6. Working phase of the genetic algorithm.
Figure 6. Working phase of the genetic algorithm.
Mathematics 11 01081 g006
Figure 7. Working phase of the ABC algorithm.
Figure 7. Working phase of the ABC algorithm.
Mathematics 11 01081 g007
Figure 8. Process of particle swarm optimization Algorithm.
Figure 8. Process of particle swarm optimization Algorithm.
Mathematics 11 01081 g008
Figure 9. Development of the research field regarding dimension reduction methods for high-dimensional gene expression data.
Figure 9. Development of the research field regarding dimension reduction methods for high-dimensional gene expression data.
Mathematics 11 01081 g009
Figure 10. Number of publications and citations for each nature-inspired algorithm in the main publishers regarding research of gene expression data research.
Figure 10. Number of publications and citations for each nature-inspired algorithm in the main publishers regarding research of gene expression data research.
Mathematics 11 01081 g010
Figure 11. Implementation challenges of nature-inspired algorithms.
Figure 11. Implementation challenges of nature-inspired algorithms.
Mathematics 11 01081 g011
Figure 12. Hybridization of nature-inspired algorithms with a ML model to optimize cancer classification and prediction performance.
Figure 12. Hybridization of nature-inspired algorithms with a ML model to optimize cancer classification and prediction performance.
Mathematics 11 01081 g012
Table 1. Comparison of different dimension reduction techniques.
Table 1. Comparison of different dimension reduction techniques.
Study TypesAdvantagesDrawbacksRegular SearchEffectiveness
Filter MethodAutonomous
Neglects classification
Quick calculation times
No communication with the classification model
Disregards important features
Univariate
Multivariate
Quicker than other feature selection techniques
lowers the feature’s significance
Wrapper MethodDependent on feature
Computerized and time consuming
Interacts with the classifier and feature selection
Overfitting risk
Dependent on the classifier
Lengthy calculation time
Time that is complex exponential overfitting risk
Dependent on the classifiers
Deterministic
Stochastic
Superior to the filter method
Extraordinary efficiency
Embedded MethodHaving a low chance of overfitting
Interacting with the classifier
Using the best FS method with the classification model
Classifier depends on the technique of selection
Propensity for overfitting
Integrated model
Simplified model
Computations are less expensive than wrapping
Feature ExtractionHigher discriminating power
Control over fitting problem
Loss of data interpretability
The transformation may be
expensive
PCA, Linear
discriminant analysis,
ICA
Used in an effective way in the hybrid algorithm
Hybrid MethodCombines multiple feature
selection and extraction techniques.
Time-consuming and challengingSearching in depth
Ideal FS
Complexity
Reduced mistake
Table 2. Commonly used algorithms derived from nature for biomedical data.
Table 2. Commonly used algorithms derived from nature for biomedical data.
AlgorithmIntroduced ByYearInspired By
Genetic AlgorithmJohn Holland1960Process of natural selection
Ant Colony OptimizationMarco Dorigo1992Foraging behavior of natural ants
Particle Swarm AlgorithmKennedy and Eberhart1995Social behavior of birds
Artificial Bee ColonyKaraboga2005Intelligent foraging behavior of bees
Fire fly AlgorithmXin-She Yang2008Flashing behavior of fireflies
Cuckoo SearchYang and Suash deb2009Obligate brood parasitism
Bat AlgorithmXin She yang2010Echolocation behavior of microbats
Whale AlgorithmMirjalli and Lewis2016Hunting mechanism of humpback whales in nature
Harris HawkHeidar2019Harris hawks hunting as a group
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Yaqoob, A.; Aziz, R.M.; Verma, N.K.; Lalwani, P.; Makrariya, A.; Kumar, P. A Review on Nature-Inspired Algorithms for Cancer Disease Prediction and Classification. Mathematics 2023, 11, 1081. https://doi.org/10.3390/math11051081

AMA Style

Yaqoob A, Aziz RM, Verma NK, Lalwani P, Makrariya A, Kumar P. A Review on Nature-Inspired Algorithms for Cancer Disease Prediction and Classification. Mathematics. 2023; 11(5):1081. https://doi.org/10.3390/math11051081

Chicago/Turabian Style

Yaqoob, Abrar, Rabia Musheer Aziz, Navneet Kumar Verma, Praveen Lalwani, Akshara Makrariya, and Pavan Kumar. 2023. "A Review on Nature-Inspired Algorithms for Cancer Disease Prediction and Classification" Mathematics 11, no. 5: 1081. https://doi.org/10.3390/math11051081

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop