Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript introduces a probabilistic clustering framework for air pollution monitoring utilizing the Expectation-Maximization (EM) algorithm. This topic is pertinent to the field of sensors, and the paper is well-structured, featuring clear mathematical derivations and coherent discussions. The concept of modeling sensor alerts as a mixture of Poisson distributions to distinguish between normal and polluted zones is theoretically robust and aligns well with contemporary interests in intelligent environmental monitoring systems.

However, several areas warrant improvement before the work can be considered for publication.

1. Experimental Validation: The study relies solely on simulated Poisson data, lacking any real-world sensor measurements. Incorporating actual air quality datasets (e.g., PM₂.₅, NO₂, or O₃ readings from IoT stations) or conducting field experiments would greatly enhance the practical credibility of the proposed method.

2. Comparative Analysis: It is essential to compare your method with conventional clustering techniques such as k-means, Gaussian Mixture Models (GMM), or Bayesian clustering. Employ quantitative metrics (e.g., RMSE, AIC, BIC, or computational cost) to substantiate the claim of “high efficiency.”

3. Section Numbering: There is an inconsistency in numbering; specifically, Section 4 transitions directly to Section 8. This should be corrected.

4. Terminology: The phrase “AI-based approach” may cause confusion, as the method is fundamentally probabilistic and does not involve machine learning or deep learning. A more accurate phrasing would be “probabilistic EM-based approach.”

5. Clarity and Reproducibility: To ensure reproducibility, kindly provide simulation parameters, random seeds, or source code as supplementary material.

In summary, the manuscript holds significant potential but requires substantial revisions. By incorporating real data validation, comparative results, and making minor editorial adjustments, the paper could contribute meaningfully to the field of intelligent air quality monitoring.

Comments on the Quality of English Language

The manuscript exhibits a strong command of English, presenting ideas clearly, with accurate and consistent technical descriptions and mathematical expressions. However, several sentences, particularly in Sections 3.1 to 3.3, tend to be somewhat repetitive or overly lengthy. Revising these sentences for conciseness would significantly enhance readability.

Furthermore, the term "AI-based approach" warrants reconsideration, as the proposed method is fundamentally probabilistic and does not incorporate artificial intelligence or deep learning techniques. A more suitable term might be "probabilistic EM-based approach," which would ensure greater precision and technical accuracy.

Despite these minor concerns, the language is fluent and appropriate for publication following light editing.

Author Response

For your convenience and review, we have attached the file with Reply Letter.

 

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The authors describe a manner how to determine if a sensor is measuring in clean or polluted air, such that the sensor module can increase or decrease its reporting frequency.

This is done by fitting three parameters to a two-component mixture distribution, and then calculating the probability that an observed data point (equal to the number of critical events detected within a fixed time interval) belongs to each of the components, which represent clean and polluted air.

The study feels quite artificial to me for the following reasons:

• If an air quality sensor would send a message, it would not just send a 'high event message', but give the measured concentration as well, which contains substantial information. The modelling approach by only counting sensor messages therefore seems not very realistic to me.
• There is no real data involved at all in the study. If there would be, then the usefulness of the proposed method might be proved in a practical situation.
• The bottom line of the presented work is just fitting 3 parameters of a probability distribution to a simulated dataset using the EM-algorithm. In itself, this is not very novel and the usefulness/ realism for existing sensor networks is not convincingly described, at least for me.

The English is very well and the presentation is also clear.

Some typographical errors:

page 2: last two paragraphs are very similar, almost identical. Please merge them.

page 4: 'select nodes' -> 'selected nodes'

're-al-world' -> 'real world'

'over-lapping' -> 'overlapping'

I think equation (4) is incorrect, with the exponent z_{i,k} outside the summation over k. I think the summation should be replaced by a product, and the exponent z_ik should be the exponent of each factor.

'that each data point that each data point' -> 'that each data point' 

Formula (8): \hat theta should appear in the middle part somewhere

superscript 'new' should not be in italics in several equations

'more numerically' -> 'numerically more'

'гunlike' -> 'unlike'

 

Author Response

For your convenience and review, we have attached the file with Reply Letter.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

See attached file.

Comments for author File: Comments.pdf

Author Response

For your convenience and review, we have attached the file with Reply Letter.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thank you for submitting the revised manuscript. I would like to commend the authors for their efforts in incorporating real-world data, enhancing methodological clarity, and rectifying the structural issues identified in the previous review. The manuscript demonstrates significant improvement. However, one crucial issue remains insufficiently addressed.

The primary concern lies in the absence of a comparative evaluation. In my previous review, I requested that the authors include a comparative analysis against well-established clustering techniques—such as k-means, Gaussian Mixture Models (GMM), or Bayesian clustering—supported by quantitative metrics like RMSE, AIC, BIC, clustering accuracy, confusion rate, or computational cost. In the revised manuscript, the authors argue that such comparisons are “methodologically inconsistent” due to differing model assumptions. While this explanation is noted, it does not replace the need for empirical evidence. In the fields of data mining, clustering analysis, and probabilistic modeling research, comparative baselines are a standard expectation. Even when methods rely on different statistical assumptions, presenting empirical comparisons is vital for readers to understand how the proposed EM-based Poisson mixture model performs relative to widely accepted alternatives. These comparisons also illustrate the practical advantages of the proposed method in terms of accuracy, robustness, and computational efficiency. Indeed, numerous published studies intentionally compare models with differing assumptions to demonstrate why a new method is more suitable for a specific data type.

To uphold the scientific rigor required for publication, the manuscript should still incorporate at least some form of empirical comparison. This could entail applying baseline clustering methods to the same synthetic or real-world dataset, reporting misclassification rates or clustering purity, providing model selection metrics such as AIC or BIC to justify the appropriateness of the Poisson mixture model, or comparing computational time to validate the claim of “high efficiency.” Even a succinct comparison table featuring two or three baseline methods would significantly enhance the manuscript and offer readers a clearer understanding of the proposed method’s relative performance.

Comments on the Quality of English Language

The overall quality of the English language in the manuscript is suitable for academic publication. The writing is generally clear, grammatically correct, and coherent, demonstrating appropriate use of technical terminology. The sentence structure is easily readable, and the logical flow between sections is largely smooth.

However, certain areas would benefit from further refinement to enhance clarity and precision. Notably, some paragraphs are overly lengthy and dense, which can obscure key concepts. Additionally, specific terminology—particularly concerning "artificial intelligence" and related descriptions—should be made more consistent and precise. 

Furthermore, some phrasing may be unnecessarily complex and could be simplified to improve readability. In summary, although the manuscript is understandable and generally well written, a minor professional language edit would significantly enhance stylistic consistency and clarity.

Author Response

First, we want to express our thanks to the reviewer for providing valuable comments that help improve paper quality. The paper has been extensively revised according to the reviewers’ comments. Below the authors’ responses are given for individual comments mentioned by the reviewer.

Comment 1: 

The primary concern lies in the absence of a comparative evaluation. In my previous review, I requested that the authors include a comparative analysis against well-established clustering techniques—such as k-means, Gaussian Mixture Models (GMM), or Bayesian clustering—supported by quantitative metrics like RMSE, AIC, BIC, clustering accuracy, confusion rate, or computational cost. In the revised manuscript, the authors argue that such comparisons are “methodologically inconsistent” due to differing model assumptions. While this explanation is noted, it does not replace the need for empirical evidence. In the fields of data mining, clustering analysis, and probabilistic modeling research, comparative baselines are a standard expectation. Even when methods rely on different statistical assumptions, presenting empirical comparisons is vital for readers to understand how the proposed EM-based Poisson mixture model performs relative to widely accepted alternatives. These comparisons also illustrate the practical advantages of the proposed method in terms of accuracy, robustness, and computational efficiency. Indeed, numerous published studies intentionally compare models with differing assumptions to demonstrate why a new method is more suitable for a specific data type. To uphold the scientific rigor required for publication, the manuscript should still incorporate at least some form of empirical comparison. This could entail applying baseline clustering methods to the same synthetic or real-world dataset, reporting misclassification rates or clustering purity, providing model selection metrics such as AIC or BIC to justify the appropriateness of the Poisson mixture model, or comparing computational time to validate the claim of “high efficiency.” Even a succinct comparison table featuring two or three baseline methods would significantly enhance the manuscript and offer readers a clearer understanding of the proposed method’s relative performance. Response 1:  We thank the reviewer for highlighting the need for empirical comparison. To reflect the reviewer’s comment, we have added a targeted analysis comparing our method with k-means and expanded the Performance Analysis section. Indeed, in this scenario we can make this performance comparison methodologically consistent in contrast to GMM (unsuitable for count data, Gaussian pdf, a priori non-Poisson pmf) or Bayesian methods (a separate research direction). We fully appreciate that centroids are fundamentally different in nature from distribution parameters. However, a key property of the Poisson distribution enables their use as estimators of the rate parameters. Thus, we focus on RMSE between estimated and true rates. We introduce a scenario with penalty function related to clustering accuracy as well. Also, we strengthened the motivation related to the Poisson nature of the data.   Comment 2:  Comments on the Quality of English Language The overall quality of the English language in the manuscript is suitable for academic publication. The writing is generally clear, grammatically correct, and coherent, demonstrating appropriate use of technical terminology. The sentence structure is easily readable, and the logical flow between sections is largely smooth. However, certain areas would benefit from further refinement to enhance clarity and precision. Notably, some paragraphs are overly lengthy and dense, which can obscure key concepts. Additionally, specific terminology—particularly concerning "artificial intelligence" and related descriptions—should be made more consistent and precise. Furthermore, some phrasing may be unnecessarily complex and could be simplified to improve readability. In summary, although the manuscript is understandable and generally well written, a minor professional language edit would significantly enhance stylistic consistency and clarity. Response 2:  To reflect the reviewer’s comment, we have agreed on the use of the MDPI English correction service with Section Managing Editor.

 

Reviewer 2 Report

Comments and Suggestions for Authors

Dear authors,

Thank you for successfully addressing my concerns. I think it is sufficient for publication.

I have a few minor remarks left:

• in equation (8), I think the subscript \theta needs to be replaced by \hat \theta
• in equations (9), (10) and (11) the superscript 'new' should not be in italics
• The formula numbering is very strange. There is a number (IG), multiple equations (5), multiple equations (26) and a number (225).
• on page 3 it still says 're-al-world' instead of 'real-world'

 

 

Author Response

First, we want to express our thanks to the reviewer for providing valuable comments that help improve paper quality. The paper has been extensively revised according to the reviewers’ comments. 

Comment 1 : 
Dear authors,
Thank you for successfully addressing my concerns. I think it is sufficient for publication.
I have a few minor remarks left:
in equation (8), I think the subscript \theta needs to be replaced by \hat \theta
in equations (9), (10) and (11) the superscript 'new' should not be in italics
The formula numbering is very strange. There is a number (IG), multiple equations (5), multiple equations (26) and a number (225).
on page 3 it still says 're-al-world' instead of 'real-world'

Response 1: To reflect the reviewer’s comment, we have implemented the edits recommended by the reviewer.

We sincerely thank you for your time and valuable comments on our manuscript. 

Reviewer 3 Report

Comments and Suggestions for Authors

The paper notably improved after its revision. However, there are still some concerns needing further attention. Specifically, comments 7, 8, and 12 do not seem to be addressed in the paper, but only in the response letter. Please, include them also in the manuscript, since readers will surely appreciate it.

Author Response

First, we want to express our thanks to the reviewer for providing valuable comments that help improve paper quality.  

Comment: The paper notably improved after its revision. However, there are still some concerns needing further attention. Specifically, comments 7, 8, and 12 do not seem to be addressed in the paper, but only in the response letter. Please, include them also in the manuscript, since readers will surely appreciate it.

Response: To reflect the reviewer’s comment, the paper has been extensively revised according to the reviewers’ comments 7,8,12. Below the authors’ responses are given for individual comments 

Comment 7 : In general, it is not clear how EM model was exploited. In other words, did the Authors consider other techniques? Did they implement comparative analyses? I deem these two points are of utmost importance.

Response 7: To reflect the reviewer’s comment, we have essentially extended the performance analysis section, add new numerical results for comparative analysis with k-mean clustering. We also noted the methodological consistency of this comparison.

Comment 8: The exploited dataset, and its features, are not clear, and a detailed description is fundamental. Similarly, it is not clear which kind of sensors are considered. Indeed, many types of different sensors can be exploited to monitor air quality. At the same time, I deem the model should additionally take into account geographic information (e.g., presence of water basins or mountains, vicinity to industrial plants or forests, etc.), as well as meteorological information (e.g., the effect of rain on air quality is well known). At the moment, I deem the model is not considering many pivotal parameters, like the ones just listed.

Response 8: To reflect the reviewer's comment, we have expanded subsection 3.2. In particular, we added the last paragraph, where we explain the essence of universality.

Comment 12: Finally, the Authors must clearly state the limitations of the proposed approach.

Response 12: To reflect the reviewer’s comment, we have added a corresponding paragraph at the end of Section 4.