Performance Improvement of Decision Tree: A Robust Classifier Using Tabu Search Algorithm
, Muhammad Rashid 2,†, Hassan Tariq 1,†, Zain Ul Abideen 3,†, Saud S. Alotaibi 4 and Mohammed H. Sinky 2Round 1
Reviewer 1 Report
The authors presented a very interesting algorithm for a timely problem. The paper is well written and will be beneficial for the healthcare research community.
Author Response
The file is attached.
Reviewer 2 Report
Review for
Performance vs. Accuracy Trade-off: A Decision Tree based Custom Classifier using Tabu Search Algorithm
The level of originality of the paper is high. The method is effective.
In this paper authors used 48 sources, containing both historical and fundamental works, as well as the latest scientific research on this topic. But the literature review can be improved. Please, add the economic influence of this deposition method and cite the papers in the discussion of study:
An, J., Mikhaylov, A. (2020). Russian energy projects in South Africa. Journal of Energy in Southern Africa, 31(3). http://dx.doi.org/10.17159/2413-3051/2020/v31i3a7809
Lisin A. (2020). Prospects and Challenges of Energy Cooperation between Russia and South Korea. International Journal of Energy Economics and Policy, Vol. 10 (3). https://doi.org/10.32479/ijeep.9070
An, J., Mikhaylov, A., Richter, U.H. Trade War Effects: Evidence from Sectors of Energy and Resources in Africa. Heliyon 2020, 6, 12, e05693. https://doi.org/10.1016/j.heliyon.2020.e05693
Mishina V.Y., Khomyakova L.I. Dedollarization and settlements in national currencies: Eurasian and Latin American experience. Voprosy Ekonomiki 2020, 9, 61-79. https://doi.org/10.32609/0042-8736-2020-9-61-79
Lisin A. Prospects and Challenges of Energy Cooperation between Russia and South Korea. International Journal of Energy Economics and Policy 2010, 10, 3. https://doi.org/10.32479/ijeep.9070
The introduction is needs to revise, this section has benefit from having a clearer structure of what to expect in the paper. Furthermore, the author(s) would benefit from being more concise in their writing, as much of the content was redundant and overemphasized. While it is good practice to assume the reader has no prior knowledge of the content, a topic and/or discussion does not need to be explained over and over again if it is stated both adequately and appropriately once.
Table 1-7 is important to explore the specifics. Some conclusions contribute to the study of the problem.
Authors need to add more details on the range of simulation considered in this work should be clearly outlined within the abstract. The current statements are vague and too general to get an idea of the work that have been accomplished.
Authors need to make clear and to add more details on Figure 6. Filling rate of dataset.
The Methods section needs more writing to explain how analyses were conducted. Also, make sure what is written belongs in each section. That is an analysis. Describing where you got your data and how you did your analyses is your Methods.
Some conclusions contribute to the study of the problem. The author not formulate the problem itself – it makes impossible to analyse the contribution of the paper. The aim or the question of the paper (or even the hypothesis of the author) are formulated.
Overall, it is very clear to grasp understanding of the manuscript and content in its current state. I strongly advise using hypothesis points to articulate and/or express material in scientific writing. Publication of this piece seems likely in any reputable scientific periodical after a correction in the writing of the manuscript.
The paper possesses a proper form of well-structured and readable technical language of the field and represents the expected knowledge of the journal`s readership.
There are minor errors in English, but this does not affect the general nature of the work. The current study brings many new to the existing literature or field. For one, the author(s) seem to have a good grasp of the current literature on their topic area (i.e., recent literature and seminal texts relevant to their study is not cited/referenced).
Author Response
The file is attached.
Author Response File: 
Author Response.docx
Reviewer 3 Report
The core step of the paper is a classification method that combines decision trees and tabu search. An application with clinical data of COVID is included. The contribution of the paper would be focused on practical results since the methods employed are well known and proposed extension is simple. However, the paper might be interesting from a practical standpoint. Literal presentation of the paper has room for improvement. There is no rationale of several parts of the proposed method. The authors make some bold statements in the introduction and along the paper, but there is no evidence provided of those statements. There are several important details omitted in experimental methodology. Besides, performance evaluation is incomplete and statistical significance of the results is not demonstrated. As a result, this paper is not recommended for publication.
Some examples of the major issues of the paper are the following:
- Consider to change the title of the paper. Accuracy is a performance measure, thus, “performance vs accuracy” is confusing.
- The contributions stated in the paper are poorly formulated. Actually, the first one is the only one contribution of the paper. Performance comparison with other methods using an application is always expected from a research work. However, contribution is limited as the proposed method is a simple extension with no rationale provided about convergence, optimization, … Besides, tuning of several parameters of the proposed method is not explained.
- There are several errata, grammar and written style issues in the paper as well as organization of the sections of the paper has room for improvement. For instance:
-- All the acronyms used in the paper should be defined the first time they appear in the text, except in the abstract, e.g., “ML”. “SMOT” should be changed to “SMOTE”.
-- “Abstract, “performance oriented custom classifier”. “Performance oriented” is redundant, all classifiers are oriented performance. “Custom” is used twice in the paper, it makes no sense to use this term.
-- Introduction section is too long. Sub-section 1.1 should be extracted to a separated section, e.g., “Literature review”. Sub-section 1.2 should remain part of the Introduction.
-- Figure 2 and the text that follows this figure are redundant. In addition this is a well-known method that should only be mentioned. That figure could be eliminated.
-- Figure 3 is not informative, this is a commonly used procedure in classification. It could be eliminated.
-- Figures 5 and 6 are unreadable.
Therefore, an extensive English proofreading of the paper is required.
- Besides, particular execution times data, a theoretical complexity (computational order) for each of the implemented methods used in comparisons should be estimated. Big O notation can be used.
- There are several highlighted statements which are not demonstrated in the paper, e.g., “it can handle more dimensions than other classification algorithms.” There is no theoretical or practical demonstration of that.
- Section 4.2.1, “False” and “True” are not categorical data, but logical data.
- Experimental section has several issues. For instance:
-- Equations (2) to (6) could be eliminated. They are very well known.
-- An analysis of receiver operating characteristic (ROC) curves should be implemented. This allows analyzing the performance of the detectors at different false alarm levels.
-- It has been demonstrated that SMOTE is useless with very small training datasets. It should be discussed in the paper as well as other recent oversampling methods based on generative adversarial neural networks and surrogating techniques.
-- Splitting of the data into training, validation, and testing datasets is not explained.
-- The variance of the performance of the methods at cross validation experiments should be estimated and discussed. In addition, an adequate test to evaluate the mean and standard deviation of the results should be implemented. This would provide statistical significance evaluation of the results.
Author Response
The file is attached.
Author Response File: Author Response.docx
Round 2
Reviewer 3 Report
The quality of the paper has been improved significantly. In general, all my concerns have been adequately addressed in the revised version of the paper. The authors have made a great effort considering the short time for paper modifications. The following subjects have been improved: title of the paper; literal presentation of the paper including written English, acronyms, and figures; experimental methodology; discussion on class imbalance; and evaluation of performance of the methods and statistical significance. Thus, the contribution of the paper is clearly shown in this revised version. In summary, I consider the contents of the paper are publishable, but the following changes should be addressed in a revised version of the paper.
- Class imbalance when the training dataset is very small should be discussed in the paper, in this case, techniques such as SMOTE is not effective. I suggest the following recent references: https://doi.org/10.1016/j.eswa.2020.113819; https://doi.org/10.3390/sym13040612.
- Please add a figure with a comparison of the receiver operating characteristic (ROC) curves estimated from the results of the implemented detection methods. Discuss about the performance of the methods at different regimes of false alarm probability.
Author Response
The file is attached.
Author Response File: Author Response.docx
Round 3
Reviewer 3 Report
The quality of the paper has been improved significantly. The discussions on detection performance of the implemented methods using receiver operating characteristics (ROC) curve analysis and on the class imbalance issue have been improved. Therefore, the paper should be ready for publications after final editing revision.
