Thermal Comfort Prediction Accuracy with Machine Learning between Regression Analysis and Naïve Bayes Classifier

Round 1

Reviewer 1 Report

General Comments: Determination of the analytical method is essential to determine an accurate thermal comfort model. The purpose of this study was to compare the method of multiple linear regression analysis and Naïve Bayes in making an accurate thermal comfort model.

 

Specific Issues:

1.       The article lack in literature review, articles need to be added to enhance readability.  Some relevant articles are mentioned below

A.      10.1109/ACCESS.2020.2985036

B.      doi.org/10.1016/j.rser.2022.112704

C.      doi.org/10.1016/j.ijhydene.2017.05.247

D.      doi.org/10.1016/j.buildenv.2019.01.058

E.       doi.org/10.1016/j.buildenv.2022.108890

2.       Tables are taking lots of space and many tables must be shifted in appendix. Just keep very relevant tables, maximum 3 to 4 tables in main text body of manuscript.

3.       Figure 2, data measuring is not as per ASHARE standard. Need to show proper orientation and sensor placement while performing experiment.

4.       Again, many spaces of manuscript are used not wisely such as variable explanation, need to shift in tabular form, before introduction section.

5.       How subjects/occupants were selected?

6.       Documents filled by the occupant’s must be attached in appendix.

7.       Discussion is the weakest part of this paper, must be enhance by doing comparative analysis. How data are varying, how weightage of data were given, and how issue of missing values were rectified must be clearly discussed in discussion.

 

8.       Author must need to include result of PMV and PPD, based on ASHARE standards.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

The article provides results on experimental analysis of thermal comfort by comparing regression analysis and Naive Bayes to predict the Thermal Sensation from data collected in a group of persons by considering 8 independent variables. The approach is very scholastic without a remarkable originality. The text is very poor dedicating more than 60% to tables of input data and percentages, schemes and pictures.

It should be necessary to better explain the methodology, check it in more cases, make a more efficient graphic synthesis of the results, keep just a table generalizing data input and move eventually the numerical table in a appendix.

Modifications are suggested as follows:

- page 4, from line 143 to 149: the text should be substituted by a Table 

- page 5, Table 1: explain in the text all statistical output (i.e. "Sig." is for sigma=standard deviation) and its meaning (also through references) for this particular study also emphasizing the expected values that correspond to strong correlation or the opposite. The aim is to guide the reader to comprehend the results shown and do not limit just to show the result in itself without further explications. Also clarify the variable X_i to which physical variable corresponds (i.e. i=7 for "humidity") by adding a column for items "i"

- page 6, line 205-221: arrange these results in a Table instead of multiple repetitions in the text: Table 3 should be sufficient with a previous explication of the application of eq. (1)

- page 7, Table 4: it should be organized differently and avoid to put twice the column indexes "-3 -2 ...2 3" but just once, same for other tables until Table 14 where it is sufficient only the probability value by previously explaining the operation of obtaining ratio on the total number of items. All the tables should find place in a Appendix and the text should better explain how have been chosen the histogram bin (i.e. equal to unity) for the different variables and a different choice could affect the results

 

Mistyping / mistakes:

- page 3, line 118: equation "Y = ...": it is more common to put the progressive number "i" of the coefficient beta_i and variable X_i as a subsript with a smaller character: please update it

- page 5, line 147-148: the symbol for Celsius degrees should be always as superscript "ºC" instead of "oC"

- page 6, line 205-221: please correct for class n: "P(n)" instead of constant "P(-3) (i.e. P(-2) instead of P(-3) at line 206 etc.)

- page 10, Table 9: pelase rearrange the "globe_temp" histogram bin from the smallest to the highest in the correct sequence

- page 12, Fig. 4: correct legend: "Linear" instead of "Linier" 

- page 12, line 297, Table 14: correct title: "Linear Regression" instead of "Regresi Linear" 

Further comments:

English language to be further reviewed.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

This manuscript aims to compare the method of multiple linear regression analysis and Naïve Bayes on building accurate thermal comfort models. The experimental results show that Naïve Bayes is superior.

 

Comments:

1. A Related Work section should be added.

2. The contributions of this work shoule be highlighted.

3. Equations should be reformatted.

4. A detailed description of the data set should be given in Section 3.

5. The work is straightforward and lack of novelty. Some remarks should be given to highlight the novelty.

6. A Threats to validity section should be added.

7. There are many other machine-learning approaches that can be used to solve the concerned problem. Why not try other approaches?

8. According to the results reported by one very recent study [1], for machine-learning-based approaches, the results might not converge to stable values. That is, the average result computed over one 10-independent run is always not equal to the result computed over another 10-independent run. For example, the result of one 10-independent run is 3.1415, and that of another run is 3.1425. It seems that the results converged to 3.14 since 3.14 is fixed in the two rounds of runs. Thus, 0.01 can be viewed as the precision level, and the comparison at the 0.01 precision level is meaningful. If someone is not aware of such a precision level problem and compares 3.1415 and 3.1425 w.r.t. their absolute values, he/she might conclude that 3.1425 is better than 3.1415 if larger is better. But if we are aware that the two values only can be compared at the 0.01 precision level (the numbers after 0.01 are meaningless), then we might conclude that the two values are roughly equivalent. In this sense, I think the precision level is very important. One recent work [1] pointed out such a precision level that exists in the existing machine-learning-based approaches. For details, please refer to Sections 3.3 and 4.3 in Ref. [1].

[1] https://ieeexplore.ieee.org/document/9733807

https://doi.org/10.1109/TSE.2022.3140599

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The authors incorporated the suggestions in the revised manuscript.

Author Response

Thank you very much

Reviewer 2 Report

The article provides results on experimental analysis of thermal comfort by comparing regression analysis and Naive Bayes to predict the Thermal Sensation from data collected in a group of persons by considering 8 independent variables. The approach still very scholastic without a remarkable originality. The text has not really increased its quality still dedicating entire pages to flowchart, screen-shots, tables or very big graphs.

It should be necessary to better explain the methodology, check it in more cases, make a more efficient graphic synthesis of the results, keep just a table generalizing data input and move eventually the numerical table in a appendix.

Modifications are suggested as follows:

- page 7-9: eliminate all the screen shots: it is note perceived as an additional proof, on the contrary: the results shown should be summarized in table in the appendix

Mistyping / mistakes:

- page 6, line 215: equation "Y = ...": should be re-written as "Y=Sum_i(beta_i*X_i)" by specifying X_i: (1: gender, 2: age,...): please come back to unknown X_i and avoid text for the unknown as in programming style

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

I think the authors did not resolve all the issues that I was concerned about in the last round of review if I did not miss some important information.

I think the authors should point out the exact line number of their revisions to help reviewers locate their revisions to a specific comment.

But I am glad to give the authors another chance to revise their work.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 3

Reviewer 2 Report

The article provides results on experimental analysis of thermal comfort by comparing regression analysis and Naive Bayes to predict the Thermal Sensation from data collected in a group of persons by considering 8 independent variables. 

Modifications are suggested as follows:

- page 4, line 170, eq.1 / page 7, line 235: write the equation in symbolic format as the first equation in annexed Word document. The variables can be written by putting the index "i" as subscript (as the second expression in annexed Word document. 

Mistyping / mistakes:

- page 7, line 220 / page 7, line 235: equation "Y = ...": the symbol "Sum()" is not necessary, the correct expression should be just as it follows: "Y = -8,796 + (-0,308)*X1 +....+1,420*X8" where the parenthesis for positive numbers can be avoided. The variables should be written by putting the numerical index "i" as subscript (as the second expression in annexed Word document.

Comments for author File: Comments.docx

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

Thanks for considering my comments. The paper now can be accepted.

Author Response

Thank you very much