Towards Fire Prediction Accuracy Enhancements by Leveraging an Improved Naïve Bayes Algorithm

Round 1

Reviewer 1 Report

Dear Authors,

The manuscript is clear, and well-written overall. The manuscript has sufficient originality, and undertaken problem is of practical. Although the results presented in the manuscript seem promising and overall approach is contributing in the body of the literature somehow, I encourage the authors to please consider the attached file suggestions to improvise the presented work more prior to its publication. Thanks

Comments for author File: Comments.pdf

Author Response

Response to Reviewer 1 Comments

 

We thank the reviewers for taking the time to evaluate the paper. The comments are shown below followed by our replies in bold text.

Reviewer #1:

The concepts, methodology, and results presented in the manuscript seem reasonable and overall approach is contributing in the body of the literature somehow. The manuscript has sufficient originality and undertaken problem is practical. Meanwhile, I encourage the authors to please consider the following few comments/suggestions carefully to further improve the presented work.

Major Comments/Suggestion

1) The title of the manuscript needs refinements. For instance, it can be modified to ‘Towards Fire Prediction Accuracy Enhancements Leveraging Improved Naive Bayes Algorithm’. Authors can refine more based on the subject matter.

Response: The title now has been refined as ' Towards Fire Prediction Accuracy Enhancements Leveraging Improved Naive Bayes Algorithm’.

 

2) In Section I, it is of paramount to highlight the limitations of the existing studies, and how proposed method address them.

Response: In Section I, limitations of the existing studies have been provided, and how have been highlighted. Existing methods including threshold design method, image recognition method, intelligent artificial-based algorithms have been discussed. For the threshold design method, the prediction accuracy is largely dependent on the threshold value. In a complex environment, false or missed alarms could be induced if the threshold is not properly designed. As for the image based method, it is delicate to environmental disturbances, such as fog, strong light, etc.. Also, for environments with little image difference, feature extraction and observation are usually difficult. For the intelligent artificial-based method, it usually requires a large number of parameters, such as the initial values of network topology, different weights, and thresholds. Properly design and collection of such parameters are difficult. Also implementation of the neural net-works require large amount of calculation and usually result in very high hardware cost. In the existing Naive Bayes (NB) algorithm, the characteristic attributes and the attribute values are not considered, which has critical effects on the classification accuracy. To improve the fire prediction accuracy, a double weighted Naive Bayes with compensation coefficient (DWCNB) method is proposed. Effectiveness has been verified via both simulation and experimental results.

 

3) The results are complete and well summarized. I would suggest adding some information about the complexity of the operations performed in the proposed platform.

Response: The contribution of this paper is that we have proposed an improved NB method for the fire prediction purpose. The influence of characteristic attributes and the attribute values that have critical influence of prediction accuracy have been considered in our platform. To properly design the attributes and the values, a 5 level orthogonal testing method is employed to properly design the coefficient in the platform. Complexity of algorithm implementation and hardware design are relatively simple compared with intelligent artificial-based platform. However, we need to run the 5 level orthogonal test to determine the coefficients, and this requires some work to set up the whole platform. Related explanations now have been added before Figure 3 in the revised manuscript with bold text.

 

4) Vector in line #: 104 has no values in it. Authors need to provide clear values of the vectors with brief description of them.

Response: This typo now has been corrected highlighted with bold text. The sample library is described with a three-dimensional vector group {}, where r represents the sample number, s is the number of the characteristic vectors, and each characteristic vector has t values.

 

5) Paper organization is missing. It is better to concisely provide the organization in Section I od the revise work.

Response: Statement about paper organization now has been added in Section I. This paper is organized as follows. Principle of Naive Bayesian classifier is discussed in section 2. Improvement of Naive Bayesian algorithm is demonstrated in section 3. In section 4, platform implementation and simulation comparisons have been dis-cussed. Experiments and conclusions are given in section 5, and 6, respectively.

 

6) In section II or IV, one comprehensive figure with input data, intermediate steps, and final output is required to clearly emphasize the method. The figure 1 is not in sufficient detail with actual values. Authors can possibly cover more details in that figure.

Response: Figure 1 now has been revised and more detail has been added.

 

7) In table 6,7,8, it is better to provide the citation of the methods with whom authors compared their results.

Response: In table 6,7,8, we didn’t use any specific NB algorithm from the citations. What we followed was the general Naive Bayes classifier algorithm.

 

8) The limitations of the presented study are not provided by the authors. It would be better to highlight them in the revised work.

Response: The limitations now have been added in section 5. We need to run the 5 level orthogonal test to determine the coefficients, and this requires some work to set up the whole platform.

 

9) A future course of action will be imperative in the revised work.

Response: Statements of future course of action now has been added in the conclusion part. In our study, different materials including wood, cotton rope, polyurethane plastic and ethanol are selected as the combustion materials. In future, more different materials could be used to verify the method. Also, it would be interesting if there are any other strategies to determine the compensation coefficient used in the prior probability calculations.

 

10) Formalization can be provided in a separate table with all symbols’ brief details.

Response: A new section ‘Nomenclature’ now has been added under the conclusion part.

 

Minor Commetns/Suggestion

1- Contribution can possibly be written with bullets/numbers concisely.

Response: Contributions are now numbered in the conclusion part.

 

2- Some more and pertinent keywords can be added related to subject matter presented in this paper.

Response: Keywords now are revised as ‘Fire prediction; Double weighted Naïve Bayes; Characteristic attributes; Experimental combustion material; 5 level orthogonal design’.

 

3- All symbols can be written in math mode for better readability of the manuscript.

Response: All symbols and formulae in the texts are now written in the math mode.

 

Reviewer 2 Report

Dear Authors,

I enjoyed reading your paper. However, it still needs interventions on the text, language and methodology used.

For instance you should include:

• references to the scientific literature when using formulas;
• text format:
  • include a keyword/expression regarding the experimental part at at Keywords section;
  • do not end you sections/subsections with a formula;
  • five methodological steps are glued to the caption/title of the Fig.1;
• extensive English language editing (for instance, only for the abstract, Grammarly reports an overall score of 76 points from 100, which means you have to improve the language for the entire manuscript; another example is "..Naïve Bayesian attributes are independence and equal importance.." which needs obvious adjustments!!! Check also the occurrences of the "Naïve" consecrated word. I also observed some with "Naive"!!!);
• proper details regarding the amount of data used (is not correct to report "..12 tests and 4,972 fire data are selected.." because of the confusion it creates: Is it about 4,972 records? Or datasets?);
• lack of multi-collinearity tests (e.g. Variance Inflation Factor/VIF values) and missing details regarding the matrix with predictors' correlation coefficients in order to check the basic assumption of independence when using even an improved Naïve Bayes technique.

Author Response

Response to Reviewer 2 Comments

We thank the reviewers for taking the time to evaluate the paper. The comments are shown below followed by our replies in bold text.

 

Reviewer2

Dear Authors,

I enjoyed reading your paper. However, it still needs interventions on the text, language and methodology used.

For instance you should include:

1) references to the scientific literature when using formulas;

Response: Proper references now have been added when using formulas.

 

text format:

2) include a keyword/expression regarding the experimental part at at Keywords section;

Response: The keywords now have been revised as ‘Fire prediction; Double weighted Naïve Bayes; Characteristic attributes; Experimental combustion material; 5 level orthogonal design’.

 

3) do not end you sections/subsections with a formula;

Response: Proper statements now have been added after the formula to avoid this problem.

 

4) five methodological steps are glued to the caption/title of the Fig.1;

Response: Statements and explanations are now added to clarify relationship between the Fig.1 and the five methodological steps.

In Figure 1, details of the numerical implementation and calculation procedures are described by the following five methodological steps:

 

5) extensive English language editing (for instance, only for the abstract, Grammarly reports an overall score of 76 points from 100, which means you have to improve the language for the entire manuscript; another example is "..Naïve Bayesian attributes are independence and equal importance.." which needs obvious adjustments!!! Check also the occurrences of the "Naïve" consecrated word. I also observed some with "Naive"!!!);

Response: English writing now has been carefully examined. Naive has been corrected, and description of ‘independence’, ‘importance’ have been corrected. Writing errors in the introduction, simulation and experiments parts also have been carefully corrected.

 

6) proper details regarding the amount of data used (is not correct to report "..12 tests and 4,972 fire data are selected.." because of the confusion it creates: Is it about 4,972 records? Or datasets?);

Response: Proper explanations of the 12 tests and the data set now have been added.

To run the verifications, the data set from the National Institute of Standards and Technology Report (Fire Research Division) [22] are selected, including 12 tests and 4,972 fire data. The selected 12 tests are corresponding to 4 different materials including wood, cotton rope, polyurethane plastic and ethanol. In each test, all the three status of OF, SF, and NF have been included.

 

7) lack of multi-collinearity tests (e.g. Variance Inflation Factor/VIF values) and missing details regarding the matrix with predictors' correlation coefficients in order to check the basic assumption of independence when using even an improved Naïve Bayes technique.

Response: The method of multi-collinearity test and analysis of correlation coefficient can indeed intuitively reflect the independence of predictor variables. In our study, the modeling starts from the final results, using simulation analysis and actual experiments to prove the superiority of the improved model, thus reflecting that the improved method is feasible when used to improve the independence and importance of naive Bayes variables.

 

Round 2

Reviewer 1 Report

Dear Authors, the paper is well organized and much improved now. My most of the concerns have been addressed. In my opinion, the paper is ready for publication after some minor edits. I acknowledge and congratulate the authors for their significant efforts and the time they spent on the revision of the Manuscript. The minor concerns that need correction before publication of this manuscript are given below.

• Please make the abstract a bit more concise if possible, by retaining just main key points of the proposed study, and main assertions only.
• Please provide more details about the organization of this manuscript in terms of sections names. For example, section 3 presents the proposed model or approach etc. In current form there is no description about authors proposal.
• In table 6, it is better to provide the citations of the methods with whom authors compared their results.
• Conclusion can also be made more concise by retaining key things only.
• Contribution can be marked in introduction section with bullets. I conclusion, it can be written in plan.
• The usage and scenarios of results utilization should be discussed with more details.
• Language problems are also there in numerous parts of the manuscript.
• In my opinion section III name is not proper, it can possibly be modified to, ‘The practical refinements/enhancements in the Naïve Bayesian Algorithm to augment fire prediction accuracy’ something like that.

Author Response

Response to Reviewer 1 Comments

Point 1: Please make the abstract a bit more concise if possible, by retaining just main key points of the proposed study, and main assertions only.

Response 1:The abstract now has been revised to be more concise.

 Point 2:Please provide more details about the organization of this manuscript in terms of sections names. For example, section 3 presents the proposed model or approach etc. In current form there is no description about authors proposal.

Response 2: More details now have been added to introduce the paper organization.

This paper is organized as follows. The principle of Naïve Bayesian classifier is discussed in section 2. The proposed model and the improvements of Naïve Bayesian algorithm are demonstrated in section 3. In section 4, the platform implementation and the simulation comparisons between different methods have been discussed. Experimental verifications and conclusions are given in section 5, and 6, respectively.

Point 3: In table 6, it is better to provide the citations of the methods with whom authors compared their results.

Response 3:Proper citations now have been added after equation (10). We follower the procedure introduced in [19] to develop the NB classifier. Weighted Naïve Bayes methods have been discussed in [16] and [17], and have been successfully used in text classification. The DWNB method is developed by using the similar idea for the fire prediction purpose. The fire characteristic attributes and attribute values are both weighted. In the proposed DWCNB method, the compensation coefficient is calculated according to the procedures demonstrated in Section 3.2.

 Point 4:Conclusion can also be made more concise by retaining key things only.

Response 4:  Conclusions now have been revised to be more concise.

Point 5:Contribution can be marked in introduction section with bullets. I conclusion, it can be written in plan.

Response 5: Contributions now have been added in introduction with bullets, and conclusions have been revised.

 Point 6:The usage and scenarios of results utilization should be discussed with more details.

Response 6: More details of usage and scenarios of results utilization have been added at end of section 5.

Point 7:Language problems are also there in numerous parts of the manuscript.

Response 7: The entire manuscript now has been checked and corrected by MDPI. We have attached the English Editing Certificate.

 Point 8:In my opinion section III name is not proper, it can possibly be modified to, ‘The practical refinements/enhancements in the Naïve Bayesian Algorithm to augment fire prediction accuracy’ something like that.

Response 8: The name of section 3 now has been revised as ‘3. The Practical Enhancements in the Naïve Bayesian Algorithm to Augment Fire Prediction Accuracy’

Reviewer 2 Report

Dear Authors,

You corrected many issues, but there are still some to solve, namely:

• multicolinearity issues in your data sample (the maximum absolute values for the correlation coefficients just for the predictors >0.5 - see the capture at https://tinyurl.com/54atfz5c which indicates a moderate correlation between your predictors - https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3576830/  ).
  That will require reporting the maximum absolute value/values of the correlation coefficients of predictors for the entire dataset/samples you used in order to avoid offering a false imagine with high accuracy scores but for inflated models;
• still not solved that unclear expression "4,972 fire data".
  What is it actually about ??? Data records / Data sets???
  That will require clarification.
• still not solved the English language and style issues: 6 correctness issues, and 14 advanced suggestions for the Abstract ( https://tinyurl.com/swn9zsvy ), 4 correctness issues, and 9 advanced suggestions for the Conclusions ( https://tinyurl.com/utkmwwrh ), 7 correctness issues, and 11 advanced suggestions for 5.Experimental Verification ( https://tinyurl.com/3cpn6258 ).
  These will require checking your entire manuscript and provide a Grammarly report with an overall score of more than 90 from the maximum of 100.

Sincerely,
D.H.

Author Response

Response to Reviewer 2 Comments

Point 1: multicolinearity issues in your data sample (the maximum absolute values for the correlation coefficients just for the predictors >0.5 - see the capture at https://tinyurl.com/54atfz5c which indicates a moderate correlation between your predictors - https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3576830/  ).

That will require reporting the maximum absolute value/values of the correlation coefficients of predictors for the entire dataset/samples you used in order to avoid offering a false imagine with high accuracy scores but for inflated models;

Response 1:We noted that the correlation coefficient of the selected dataset shown in table 1 exceeds 0.5. Due the page limit, we didn’t put all the 4972 dataset in the manuscript. Now we have calculate the correlation coefficient matrix for the entire 4972 dataset used in the study. The value is [1.0000, 0.4795, 0.2598; 0.4795, 1.0000, -0.0214; 0.2598, -0.0214, 1.0000]. The maximum absolute value is 0.4795.

 Point 2:still not solved that unclear expression "4,972 fire data".

What is it actually about ??? Data records / Data sets???

That will require clarification.

Response 2: In table 1, we have 15 data set. In each data set, there are three data and a corresponding fire category. The three data representing the values of temperature, smoke concentration and carbon monoxide concentration. There are entire 4972 such data set used in the study, selected from 12 tests corresponding to different combustion materials. Proper clarifications now have been added in front of table 1.

Point 3:still not solved the English language and style issues: 6 correctness issues, and 14 advanced suggestions for the Abstract ( https://tinyurl.com/swn9zsvy ), 4 correctness issues, and 9 advanced suggestions for the Conclusions ( https://tinyurl.com/utkmwwrh ), 7 correctness issues, and 11 advanced suggestions for 5.Experimental Verification ( https://tinyurl.com/3cpn6258 ).

These will require checking your entire manuscript and provide a Grammarly report with an overall score of more than 90 from the maximum of 100.

Response 3: The entire manuscript now has been checked and corrected by MDPI. We have attached the English Editing Certificate.

Please see the attachment

Author Response File: Author Response.docx