One-Class SVM Model-Based Tunnel Personnel Safety Detection Technology

Round 1

Reviewer 1 Report

The article presents an interesting use-case for applied machine learning - tunnel personnel safety. Dataset preparation phase is described in great details. As a main ML model single-class SVM is selected and evaluated in various experiments.

However reviewer has several questions / unclear aspects of the research described below.

1) 

There is mentioned "LSTM neural network" in the title and the name of the 2nd section. However the description contains details only about SVM model. Reviewer would like to see implementation details of the recurrent  neural network, justification of its need, performance evaluation, etc. Moreover, why SVM model is needed if the neural network is already there? Why SVM can't be substituted by additional dense layer in the NN?

2)

Reviewer would like to see more details about the initial datasets. How many examples are there? How it is split into train/test slices. It is not very clear how early warning examples are created.

3)

Data normalisation is done using min/max approach. What is the reasoning behind this decision in a context of anomaly detection? What will happen if an anomalous previously unseen data example will appear which is outside the given range?

Line 177: The sentence is not clear, probably outputs and inputs are mixed.

Figure 7 is not informative, it is not clear how to interpret the differences between two charts.

Figrue 9 should be improved (at least grid is needed).

Tables 9-12 probably can be substituted by chart. It is not clear how the reader would use the data presented in these tables: e.g. Recall=87.69% (is it good or bad? Should reader try to compare it with the numbers in other tables?)

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

I recommend the acceptance of the paper.

Author Response

Dear Reviewer, Thank you very much for recognizing our paper and for your direct acceptance comments.

Reviewer 3 Report

Assessment report of article

Manuscript ID: applsci-2102795

Title: Study and application of tunnel personnel safety detection technology based on One-Class LSTM neural network

Dear Authors,

The paper presents significant information and analysis, which I can appreciate. Additionally, the theme of the paper is within the scope of the journal and the results look partially reliable. This is an original manuscript with average scientific value. There are an overall coherence and relation to the scope of publication in the journal. However, I cannot recommend its publication at this stage for the following major and minor points.

·       The manuscript title is too long. A shorter title would be fine

·   The introduction section should be updated with involve the recent articles

·    The discussion should be added. It is better to make a separated discussion section and discuss the result of the present study with the result of the previous studies. Additionally, how would your study help future investigations?

·       Conclusion section needs to be also more scientifically written.

Sincerely,

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

It is meaningful to make study of tunnel personnel safety detection technology. However, some concerns should be noted before the article can be published.

1. The results are more than adequate but there is little or no discussion of the results and no discussion of how the results contribute to the larger field of assessments. The conclusions do not conclude anything but rather just revamp the purpose and result of the study.

2. The introduction is too confused. The introduction needs to set up the story. The novel aspects no sufficiently treated in previous literature should be emphasized. Previous works related to safety management and risk assessment (e.g. 10.1016/j.psep.2022.10.078) will be helpful to improve the introduction of research background.

3. There is currently very little discussion of the implications and limitations of the current analysis, the authors should add some additional discussion of these issues.

Author Response

Dear reviewers, In response to question 1, we thank you very much for your valuable suggestions and we have revised the conclusion section to discuss and summarize the results of this paper in accordance with your comments.

Dear Reviewer, In response to question two, as you requested, we have introduced the introduction section and cited previous work related to safety management and risk assessment (e.g., 10.1016/j.psep.2022.10.078).

Dear Reviewers, In response to question three, we have added implications and limitations to the current analysis and have updated the article in the introduction section.

All the above updates are marked in red in the manuscript, thank you for your suggestions

Round 2

Reviewer 1 Report

Dear Authors, 

Thank you for answers.
There are no "track changes" enabled in the submitted document, hovever I hope the changes are incorporated into the artricle.

> Figure 7: The explanation given is clear, but for my opinion the figure itself still can be improved. It is not obvious that (a) part also represents data points. I would suggest to add blue dots to part (a) along the red line, thus showing that on part (a) all samples are exactly on the distribution line, but on part (b) they have dither.

Tables 9-12:

> All indexes are around 90%, indicating good performance of the model. Our results improved by an average of nearly 10% compared to previous algorithms.

It is still not clear how reader will use all these numbers. It is hard to compare datasets/experiments/models, hard to evaluate how "far away" given record from said 90% performance.

It would be much more informative to replace these tables with bar charts showing different experiments/datasets alongside each other (easy to compare). 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

It is an original manuscript with average scientific value. This manuscript presents an interesting and beneficial study. Additionally, the theme of the paper is within the scope of the journal. The paper is technically correct, well organized. The illustrations are adequate and the results look secure. I recommend it for publication.

Author Response

Dear reviewers, thank you very much for your affirmation of our paper, and once again, our sincere thanks to you.

Reviewer 4 Report

I recommend publishing of this paper.

Author Response

Dear reviewers, thank you very much for your affirmation of our paper, and once again, our sincere thanks to you.