Method for Determining the Dominant Type of Human Breathing Using Motion Capture and Machine Learning

Round 1

Reviewer 1 Report

I believe that the article is interesting, and that the proposal has potential, but it has many aspects to improve that prevent me from recommending its publication in its current state.

An abstract should summarize the major aspects of the entire paper in a prescribed sequence, and it can include 1) the overall purpose of the study and the research problem(s) you investigated; 2) the basic design of the study; 3) major findings or trends found as a result of your analysis; and, 4) a brief summary of your interpretations and conclusions.

The motivation of the paper could be improved by including a clearer organizational structure: context, problem addressed, research questions to be solved, hypotheses, scientific method to validate the hypotheses, etc.

Why the authors have chosen training 60% and testing 40% of the data.

How you have validated the underfitting and overfitting of your model.

 

Discuss the loss function used in your model.

Author Response

I believe that the article is interesting, and that the proposal has potential, but it has many aspects to improve that prevent me from recommending its publication in its current state.

Thank you for reviewing our work and providing helpful recommendations.

An abstract should summarize the major aspects of the entire paper in a prescribed sequence, and it can include 1) the overall purpose of the study and the research problem(s) you investigated; 2) the basic design of the study; 3) major findings or trends found as a result of your analysis; and, 4) a brief summary of your interpretations and conclusions.

We have rewritten the abstract according to the prescribed sequence.

The motivation of the paper could be improved by including a clearer organizational structure: context, problem addressed, research questions to be solved, hypotheses, scientific method to validate the hypotheses, etc.

We improved the introduction section by including more description of the context and problem addressed (see lines 30-47) and clearly formulated research questions (see lines 88-93 and the relevant discussion in the Discussion section).

Why the authors have chosen training 60% and testing 40% of the data.

We added the discussion of the splitting the sample (see line 182-191)

How you have validated the underfitting and overfitting of your model.

We added the graph of the loss functions and discussion of overfitting (see lines 421-423).

Discuss the loss function used in your model.

We added the discussion of the loss function (see lines 413-420)

Reviewer 2 Report

The paper addresses an interesting topic and the novel contribution seems good. Please find below some points that still need to be improved before reconsidering the manuscript for publication:

1) In the Introduction, authors should provide a short but more general discussion concerning the use of different types of wireless sensors and sensing devices in the emerging IoT era. This would better motivate how far can we now go with these new technologies, based on wireless sensor networks, in improving or either revolutionising various contexts of the human society.  Prominent examples are, besides the medical aspects, also timely topics such as environmental monitoring, privacy and security, etc. In my opinion, this initial part should be thus enriched. I would suggest some interesting and quite recent references that could be considered:

- "Toward integrated large-scale environmental monitoring using WSN/UAV/Crowdsensing: a review of applications, signal processing, and future perspectives", Sensors, 2022.

- "Security and privacy in wireless sensor networks: Advances and challenges", Sensors, 2020.

2) At the end of the Introduction, please provide a comparative table where existing approaches are categorised based on their advantages/disadvantages. This would help to better clarify the main novelty of the present contribution.

3) Could you discuss why in Fig. 18 there is an increasing trend only when going from a time window of length about 6/7 to a time window of about 10/11? It seems counterintuitive since the subsequent trend as a function of the length of the time window is decreasing.

4) The results reported in Table 7 could be corroborated by a complexity analysis, so as to highlight potential trade-offs between the achieved classification accuracy and the associated computational cost.

Author Response

Thank you for reviewing our article and providing helpful recommendations.

1) In the Introduction, authors should provide a short but more general discussion concerning the use of different types of wireless sensors and sensing devices in the emerging IoT era. This would better motivate how far can we now go with these new technologies, based on wireless sensor networks, in improving or either revolutionising various contexts of the human society.  Prominent examples are, besides the medical aspects, also timely topics such as environmental monitoring, privacy and security, etc. In my opinion, this initial part should be thus enriched. I would suggest some interesting and quite recent references that could be considered:

- "Toward integrated large-scale environmental monitoring using WSN/UAV/Crowdsensing: a review of applications, signal processing, and future perspectives", Sensors, 2022.

- "Security and privacy in wireless sensor networks: Advances and challenges", Sensors, 2020.

We enhanced the introduction according to your suggestions (see lines 48-55).

2) At the end of the Introduction, please provide a comparative table where existing approaches are categorised based on their advantages/disadvantages. This would help to better clarify the main novelty of the present contribution.

We provided the table you suggested (see line 87, Table 1)

3) Could you discuss why in Fig. 18 there is an increasing trend only when going from a time window of length about 6/7 to a time window of about 10/11? It seems counterintuitive since the subsequent trend as a function of the length of the time window is decreasing.

We added the graphs of mean accuracy and its standard deviation depending on the time window and provided extensive discussion on choosing the time window (see pages 19-21).

4) The results reported in Table 7 could be corroborated by a complexity analysis, so as to highlight potential trade-offs between the achieved classification accuracy and the associated computational cost.

We added training time and working time to the table (now Table 8, page 22) to evaluate the computational cost. 

Round 2

Reviewer 1 Report

can be accepted 

Reviewer 2 Report

The authors have satisfactorily addressed all my comments and improved the manuscript accordingly. I recommend acceptance of the manuscript.