Next Article in Journal
Deep Clustering of Student Code Strategies Using Multi-View Code Representation (CMVAE)
Next Article in Special Issue
Effects of Airborne Particulate Matter in Biomass Treatment Plants on the Expression of DNA Repair and IL-8 Genes
Previous Article in Journal
Analysis of the Possibility of Using 3D Printing for Emergency Replacement of Damaged Elements in the Marine Industry
 
 
Article
Peer-Review Record

Explanation of Air Quality Data Using Takagi–Sugeno Fuzzy Inference System

Appl. Sci. 2025, 15(7), 3461; https://doi.org/10.3390/app15073461
by Alžbeta Michalíková 1,2
Reviewer 1: Anonymous
Reviewer 2: Anonymous
Appl. Sci. 2025, 15(7), 3461; https://doi.org/10.3390/app15073461
Submission received: 26 February 2025 / Revised: 18 March 2025 / Accepted: 19 March 2025 / Published: 21 March 2025
(This article belongs to the Special Issue Advances in Air Pollution Detection and Air Quality Research)

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper presents an interesting approach to air quality data explanation using Takagi-Sugeno Fuzzy Inference Systems . The author addresses the need for explainability in data science and demonstrate how FIS can be used to create IF-THEN rules that describe the relationship between input and output variables. 

1,Two examples of  membership functions of figures 1 and 2, please use formulas, not words to express  these membership functions.

2,Mamdani FIS is used?  It appears in Page 6. Is this the method of comparative emphasis? If so, there should be a description following it in section 4,5 or 6. Please highlight the advantages and disadvantages of Mamdani FIS and TS FIS.

3, what is the difference between 12 rules and 14 rules.  add 2 new rules or establish 14 new rules?Providing more details about the implementation.

Author Response

I would like to thank the reviewer for their comments and suggestions which clearly improved the quality of the article.

Comment 1: Two examples of membership functions of figures 1 and 2, please use formulas, not words to express these membership functions.

Response 1: Thank you for this suggestion, the formulas for the mentioned functions were added to the description of specified figures.

 

Comment 2: Mamdani FIS is used? It appears in Page 6. Is this the method of comparative emphasis? If so, there should be a description following it in section 4, 5 or 6. Please highlight the advantages and disadvantages of Mamdani FIS and TS FIS.

Response 2: Thank you for the question and comment. The Mamdani FIS was not used in the experiments. This type of FIS was mentioned because of its shortcomings which motivated the use of TS FIS. The disadvantages of Mamdani FIS were highlighted in the revision of the text.

 

Comment 3: What is the difference between 12 rules and 14 rules. Add 2 new rules or establish 14 new rules?Providing more details about the implementation.

Response 3: Thank you for your question. We did not create a new rule base of 14 rules, new two rules were added to the already existing rule base. The process of adding two rules was highlighted in the revision of the article.

Reviewer 2 Report

Comments and Suggestions for Authors

Some comments are given.

  1. The authors replaced the output linear functions with constant functions and subsequently optimized the system to achieve the lowest approximation error. This performance improvement requires further explanation.
  2. According to the structure of the fuzzy logic system in the text, the scalability of the system is very poor. For prediction problems in other regions, the effectiveness may be poor.
  3. Many images seem to be obtained directly from screenshots, and the quality needs to be improved.
  4. This phenomenon occurs because ANFIS is not a classical neural network. This description is confusing and requires further explanation.

 

--------------------------------------------------------------------

Additional Comments:

1 The authors replaced the output linear functions with constant functions and subsequently optimized the system to achieve the lowest approximation error. This performance improvement requires further explanation.
2 According to the structure of the fuzzy logic system in the text, the scalability of the system is very poor. For prediction problems in other regions, the effectiveness may be poor.
3 Many images seem to be obtained directly from screenshots, and the quality needs to be improved.
4 This phenomenon occurs because ANFIS is not a classical neural network. This description is confusing and requires further explanation.
5 Please use the same database as in this article, calculate by constructing the neural network model, and compare the results with the effect of the fuzzy system in this paper, to prove the superiority of this method.
6 Please demonstrate the correlation between vehicle density and PM10 value, such as whether vehicle type (such as the proportion of fuel vehicles and electric vehicles) will affect this correlation.

Comments on the Quality of English Language

The English could be improved to more clearly express the research.

Author Response

We would like to thank the reviewer for their comments and suggestions which clearly improved the quality of the article.

From the structured part of the review, we concluded that the following changes are needed:

  • We added the related work to the introduction.
  • The design of the proposed system was described more clearly, and a flowchart of the process was added.
  • The text related to the used methods and reached results was rewritten for higher readability.

Comment 1: The authors replaced the output linear functions with constant functions and subsequently optimized the system to achieve the lowest approximation error. This performance improvement requires further explanation.

Response 1: Thank you for this suggestion. The selection of constant functions as outputs was done for the purpose of the explainability of the system, however, this change produces a slightly higher error rate when compared with the linear function systems. Therefore, the difference between the two systems considered in the text lies in the explainability itself.


Comment 2: According to the structure of the fuzzy logic system in the text, the scalability of the system is very poor. For prediction problems in other regions, the effectiveness may be poor.

Response 2: Thank you for this comment. We agree with this comment, the systems focused on air quality data are notoriously location-specific. Hence, the use of such systems in other locations is rarely possible. We tried to highlight this idea in the related work section.


Comment 3: Many images seem to be obtained directly from screenshots, and the quality needs to be improved.

Response 3: Thank you for your comment. We upscaled both screenshots from MATLAB software.


Comment 4: This phenomenon occurs because ANFIS is not a classical neural network. This description is confusing and requires further explanation.

Response 4: Thank you for the comment. We changed the description of ANFIS to obtain a clear explanation of what was meant in the original version of the manuscript.


Comment 5: Please use the same database as in this article, calculate by constructing the neural network model, and compare the results with the effect of the fuzzy system in this paper, to prove the superiority of this method.

Response 5: Thank you for this suggestion. The advantage of this system is not so much the achieved accuracy (which, incidentally, is one of the best among the methods used) but the possibility of explaining the influence of individual inputs on the considered output. Such an approach allows local governments to focus on reducing air pollution by reducing the most polluting factors.

Using a neural network model for this task would produce a black-box model, which would need additional methods to explain the decisions made by this system. Even though such a neural network-based method could produce comparable results, its training and explaining would be much more time- and memory-consuming.


Comment 6: Please demonstrate the correlation between vehicle density and PM10 value, such as whether vehicle type (such as the proportion of fuel vehicles and electric vehicles) will affect this correlation.

Response 6: Thank you for your suggestion. In the article, we examined the correlation of several pollutants at the same time. The division of vehicles into electric and gasoline is of course important, especially nowadays. Such data is not yet captured by the sensors used in the study, and therefore we do not have the necessary data for such an analysis. Also, the results of such research can contribute to municipalities’ start of monitoring of the mentioned vehicles separately.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The author has already addressed my three questions. This paper could be accepted.

Reviewer 2 Report

Comments and Suggestions for Authors

No further comments

Back to TopTop