5.3.1. Support Vector Machine (SVM)

For regression and classification, supervised learning methods used are known as Support vector machines (SVMs) [142]. They are the members of the family of generalized linear classifiers. In tasks like recognition of handwriting, SVM is very popular as it uses pixel maps as input and also gives accuracy that is comparable to sophisticated neural networks with explained characteristics [143]. Many applications are using this in tats like analysis of handwriting, recognition of face, and so on, especially on the application that is based on the classification of pattern and regression. Vapnik [144] developed the Support Vector Machines (SVM) and due to its much better empirical performance, it gained popularity. Conventional neural networks use the principle of Structural Risk Minimization (SRM) that is superior to the traditional Empirical Risk Minimization (ERM) principle [145]. Error in the training data is reduced by the ERM and SRM reduces the upper bound on the risk that is expected. SVMs are nowadays developed to simplify the problems based on the regression as well as problems that are based on the classification [146]. In the detection of waterborne pathogens [147] and cancer diagnosis [148], SVM is used extensively.

#### 5.3.2. Feedforward Artificial Neural Networks (ANN)

The model on simple biological processing of cells (Neurons) and connectivity of the brain is successfully developed through Artificial Neural Network (ANN) as it is a parallel distributed system. Furthermore, a system that is distributed in a parallel way consists of processing elements that can work in parallel and the elements are interconnected (Figure 12) [149]. In addition to this, Artificial Neural Network (ANN) is a system that serves information as distributed system coefficients with a multiple linear regression (MLR) process [150]. Generally, by training the network, information can be gained and need not be known explicitly in advance. Because of that, the adaptability that is combined with the distributed representation in the network is described as the property of recognition of similar patterns as well as the generalisation of abstract patterns within the network input space. Due to this, ANN can degrade and compensate the performance with unreliable inputs and bad training data [151]. Engineers and scientists are provided access to deep learning tools designed by institutions like Microsoft (CNTK), Pytorch, Google (TensorFlow) and universities (Theano). One of the major reasons for the performance of the ANN is significantly affected by the parameters like hidden layer size [152].

**Figure 12.** Typical structure of ANN. Reproduced from Reference [149] with permission. Copyright 2018, Elsevier.

#### 5.3.3. Convolutional Neural Network (CNN)

Analyzing images like computed tomography (CT) images, X-ray images, and magnetic resonance images is done by a very efficient deep learning type known as CNN [153–155]. Nowadays, solutions based on Artificial Intelligence (AI) are used in various biomedical complications such as detecting breast cancer, brain tumour, etc. [156]. The research communities mostly use this learning method among the different deep learning methods, as convolutional neural networks (CNNs) have provided great accuracy in the classification of the image [157]. On the other hand, there is only one article [158] that explained the categorization of viral and bacterial pneumonia and the applications of deep machine learning algorithms in detecting pneumonia is reported by various research teams [159]. For detecting pneumonia, the parameters of deep layered CNN are varied by the authors in different works. In the radiography of the lungs, interstitial or alveolar patterns of diffuse opacification are observed. Bacterial infection is proof of alveolar infiltration in the radiography of the chest especially in patients with lobar infiltrates [160]. With the enhanced performance in the classification of the images, CNN is very famous. Some important characteristics in an image such as temporal and spatial are extracted by using the convolutional layer that is present in the network. For the reduction in the computation efforts of the computer, the weight-sharing technique is applied [161]. A CNN comprises three building blocks as a max-pooling (subsampling) layer for downsampling the image and reducing the dimensionality and hence reducing the computational efforts of the computer, a fully connected layer for equipping the network with the capability to classify, and a convolution layer for gathering knowledge about characteristics [162].

## 5.3.4. Recurrent Neural Networks (RNN)

Neural Network in which the output of the previous step is used as an input for the present step is known as Recurrent Neural Network (RNN). In classical neural networks, for the requirement of the prediction of the upcoming word in the sentence, the preceding words are used; therefore, it is important to memorize the preceding words otherwise; inputs and outputs are not dependent on each other. By using a hidden layer, RNN came into being to deal with this issue. A hidden state is the foremost characteristic of RNN, which memorizes some information about the series. In studies related to sequential data, among various learning methods, RNN attracts the researchers significantly [163]. In every recurring round, RNN is compatible with sequential data and time-series as the network structure is particularly developed for representing historical information [164]. Problems based on sequence mapping are widely solved by using RNN such as reinforcement learning, handwriting recognition, speech recognition, and sequence generation because of the characteristic of propagating previous information along with time via recurring connections [165–168]. For the detection of interactions between proteins and genes, biomedical researchers are now using RNN [169]. According to the latest studies, a promising segmentation of brain tumors can be achieved by training the RNN [170]. An extraordinary type of RNN that is capable of long-term dependencies is Long Short-Term Memory (LSTM) networks [171]. For the detection of modifications in the DNA, Bidirectional RNN with LSTM was designed [172]. The accuracy of the nanopore sequencing has been enhanced with the help of algorithms that are based on RNN [173].

#### **6. Challenges and Solution**

#### *6.1. Challenges*

The electrochemical sensors are characterized by three main challenges in their proper functioning, i.e., stability and reproducibility, low LOD value, and real sample sensitivity. These are also the major challenges in the fabrication of electrochemical biosensors.


#### *6.2. Solutions*


Since it is a multidisciplinary field, the experts from diverse backgrounds such as materials science, nanotechnology, electronics, chemistry, physics, engineering, medical practitioners, working together will eventually lead to development, enhancement, and eventual commercialization of electrochemical biosensors.

#### **7. Future Outlook**

The field of electrochemical biosensors has gained immense popularity in the last few decades, and it is quite evident from the publications in this respect, as shown in Figure 13 [174]. As it observed from the figure, the number of publications in "electrochemical biosensors" has increased continuously and a similar trend is expected to be followed in the coming years. The market of the electrochemical biosensors is expected to grow at 9.7% annually and be worth 24 thousand million dollars [119]. In clinical applications, the electrochemical biosensors are used for point-of-care (POC) testing and will reach a value worth of 33 billion dollars.

As mentioned earlier, the electrochemical biosensors are effective in detecting biowarfare agents and this application will become more advanced and available to new species in the future. There is a huge scope for technical and design modifications in this regard. Another field is wearable electrochemical biosensors, and some work has already been done in this field. The wearable biosensors have opened a new market for electrochemical biosensors, and these would prove very effective in sports and athletics. The substrate based on polymers has already been developed. This provides extra flexibility and strength to the wearable biosensor. The wearable biosensors use sweat, saliva, biofluids, etc. for the detection, thus preventing the painful process of taking blood serum for analysis. Moreover, wearable biosensors can be easily installed as bandages, tattoos, contact lenses, cloth, etc. These wearable electrochemical biosensors would be helpful in the food industry, sports, monitoring animal and human health, and defence. Recently, wearable electrochemical biosensors have been developed for monitoring diabetes and this has revolutionized the market. However, the application is in its early stages and would prove to be very effective in monitoring the health of a patient more precisely. The application of nanomaterials in fabricating electrochemical biosensors has enhanced the properties of the biosensors and more such materials need to be searched in the future. The printing of an electrochemical biosensor on the body in the form of a tattoo is another new dimension that has been invented and a lot needs to be done in this regard in the future as this would help in the easy installation of these biosensors. Better biosensing materials and effective bio-fabrication techniques will become important for producing electrochemical biosensors in the future. In addition to this, POC devices that could be easily used at home for monitoring the health of a patient, detecting pathogens in water sources, etc. will be a major concern in the future. To overcome the problems with large-scale production, microelectronics would be a great help in reducing the size of the electrochemical biosensors. In the future, the electrochemical biosensors would accompany the communication technology, and this will help in remotely monitoring the health of a patient. Moreover, the quality and safety of food products can also be managed remotely with such development. The hybrid biosensors, i.e., a combination of two or more biosensors, would help monitor various health aspects. In addition to this, the search for multifunctional materials for biosensing would also provide the same functions. The current global scenario of the COVID-19 pandemic has taken the world by surprise and current testing kits are facing some issues such as time and credibility of the results. The electrochemical biosensors have been successfully used to detect viruses, bacteria, and other antigens. They provide results in less time with efficiency. Thus, electrochemical biosensors would be very helpful in detecting the COVID-19 virus and there is a great scope for developing such biosensors. Recently, degradable electrochemical biosensors have been developed using polyethyleneimine coated over MWCNTs [175]. This paves the way for developing such biosensors without affecting the sustainable development goals. The future has great scope and potential for electrochemical biosensors in various fields of applications. The new and multifunctional biosensing materials need to be searched and the bio-fabrication techniques need to be made more advanced. Microelectronics will help in reducing the size of the biosensors which eventually will be a solution to the various problems.

**Figure 13.** Recent reviews/articles published on electrochemical biosensors from the year 2000 to July 2021. Reproduced from Reference [174]. Copyright 2021, https://www.dimensions.ai/; accessed on 12 August 2021.

#### **8. Conclusions**

Biosensors are analytical devices that convert a biological response into an electrical signal. An ideal biosensor should be temperature and pH-independent, recyclable, and specific. The field of electrochemical biosensors has gained immense popularity in the last few decades, and this is quite evident from the growing number of publications. As

observed from the figure, the number of publications in "electrochemical biosensors" has increased continuously and a similar trend is expected to be followed in the coming years. The market of the electrochemical biosensors is expected to grow at 9.7% annually and be worth 24 thousand million dollars. In clinical applications, the electrochemical biosensors are used for point-of-care (POC) testing and will reach a value worth of 33 billion dollars. Biosensors have been applied in many fields namely the food industry, medical field, marine sector, etc., and they provide better stability and sensitivity as compared with the traditional methods. The development and research in electrochemical biosensors are becoming popular in biology, electronics, material science, and engineering. Applications of nanomaterials in biosensors provide opportunities for building up a new generation of biosensor technologies. Nanomaterials improve mechanical, electro-chemical, optical, and magnetic properties of biosensors and are developing towards single-molecule biosensors with high throughput biosensor arrays. The fusion of electrochemical biosensing with nanotechnology and the growing need for inexpensive, mass production of single-use biosensors promises to change the unfortunate fact that glucose test strips are the sole product of the electrochemical biosensor industry to have achieved commercial success. Machine learning helps in interpreting large sensing data and has been used for various applications. The data obtained sometimes are not clear and possess some noise and disturbance. Machine learning helps in analyzing such data and obtain the required results. In the case of biosensors, the presence of impurity affects the performance of the sensor, and machine learning helps in removing signals obtained from the contaminants to obtain a high sensitivity. Presently, electrochemical biosensors are helping in combining biology with electronics. The biosensors are becoming efficient, smaller, and cost-effective. In the future, the electrochemical biosensor will revolutionize the field of diagnosis, health care, food security, and defense.

**Author Contributions:** Conceptualization, A.K., H.F. and S.A.; methodology, A.S. (Anoop Singh), A.S. (Asha Sharma) and A.A.; validation, A.K.S. and S.A.; data curation, S.A. and A.S. (Anoop Singh); writing—original draft preparation, A.S. (Anoop Singh), A.S. (Asha Sharma) and A.A.; writing—review and editing, S.A., A.K., A.K.S. and H.F.; visualization, S.A.; supervision, A.K. and S.A.; funding acquisition, A.K. and H.F. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported in part by JSPS KAKENHI Grant Number JP17H01224, JP18H05471, JP19H01122, JST COI Grant Number JPMJCE1314, JST -OPERA Program Grant Number JPMJOP1844, JST -OPERA Program Grant Number JPMJOP1614, Moonshot Agriculture, Forestry and Fisheries Research and Development Program (MS508, Grant Number JPJ009237) and the Cabinet Office (CAO), Cross-ministerial Strategic Innovation Promotion Program (SIP), "An intelligent knowledge processing infrastructure, integrating physical and virtual domains" (funding agency: NEDO) and the Program on Open Innovation Platform with Enterprises, Research Institutes and Academia (OPERA) from JST.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest.

## **References**


MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Biosensors* Editorial Office E-mail: biosensors@mdpi.com www.mdpi.com/journal/biosensors

MDPI St. Alban-Anlage 66 4052 Basel Switzerland

Tel: +41 61 683 77 34 Fax: +41 61 302 89 18

www.mdpi.com ISBN 978-3-0365-4326-0