Individualized Stress Mobile Sensing Using Self-Supervised Pre-Training

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Manuscript Title: Personalization of Stress Mobile Sensing using Self-Supervised

Learning

General observation: The article`s title is interesting, and the problem has relevant significance. The research designs and methods have nicely covered the gap between the problem and its solutions. The results discussions are convincing. I recommend an acceptance with few minor changes as mentioned below.

 

Comments:

·         Add a summary of the related work (research gaps/challenges, opportunities) at the end of related work section with a table that compares the relevant works.

·         The formal problem statement section is missing that is supposed to be added before the methodology section, clearly indicating the motivation and research gap.

·         It would be better to represent the RMSE values of table 1 in a graph.

·         The graphs in figure 5 should be placed and explained separately corresponding to each question (1-6)

·         Several references are more than 5 years old, and paper also lacks the most recent works.

·         The list of references is too large considering the paper as an original article. I recommend to shrink it to 30-40 references at max. As it can be seen that at several places the authors have cited more than one paper for a single statement, which can be reduced to one.

Author Response

We would first like to thank the Reviewer for providing valuable feedback. We have incorporated all of the comments into our revised manuscript, and we address each comment point-by-point below:

 

“Add a summary of the related work (research gaps/challenges, opportunities) at the end of related work section with a table that compares the relevant works.”

We have included a table in the revised manuscript demonstrating the main focus and opportunities of the existing literature.

“The formal problem statement section is missing that is supposed to be added before the methodology section, clearly indicating the motivation and research gap.”

We have included the hypothesis in the introduction just before the contribution.

 

“It would be better to represent the RMSE values of table 1 in a graph”

We have displayed the RMSE scores of each subject in a graph in the revised manuscript in Figure 5.

“The graphs in figure 5 should be placed and explained separately corresponding to each question (1-6)”

We have explained the results regarding all questions in Section 4.2 in the revised manuscript.

“Several references are more than 5 years old, and paper also lacks the most recent works.”

We have removed most of the references that are more than 5 years old and also added 3 more more modern research papers. Please see reference no. 22, 26 and 27. 

 

“The list of references is too large considering the paper as an original article. I recommend to shrink it to 30-40 references at max. As it can be seen that at several places the authors have cited more than one paper for a single statement, which can be reduced to one.”

We have shortened the reference list to 34 in the revised manuscript. We have also removed multiple references for single statements.

 

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have proposed an approach to stress mobile sensing using self-supervised learning. The paper is well-written, and the flow of information is smoothly presented. I enjoyed reading it very much. I have only two remarks: please consider adding a paragraph to the first section to outline the organization of the paper, and there are several instances where acronyms are used without prior explanation; kindly provide the full set of words before using an acronym for the first time.

Author Response

We express our gratitude to the reviewer for the positive review. We have added an explanation of each acronym before using them.

Reviewer 3 Report

Comments and Suggestions for Authors

 

 

The authors, in this paper, have proposed a personalized stress prediction system using self-supervised learning based one-dimensional CNN. The proposed system is evaluated using a single data set only. The benefits of the proposed methodology are shown by comparing its performances with the similar system trained with a supervised learning in terms of RMSE only. The paper is written very well. The applicability of the proposed research in the domain of digital health care is well explained. However, there are a few issues related to description of the proposed technique, collected data set and typographical errors based on which the following suggestions are made to further improve the current version of this paper.

1   The abbreviated terms, e.g., ‘EDA’, ‘HRV’ must be expanded on their first use.

2  In Section 3, data samples used for training the proposed model should be illustrated for its better understanding.

3.     Author should mention the name of the learning algorithm used for supervised learning of the proposed technique and also specify the reason for adding one new FC layer in the fine-tuned pretrained model based on self-supervised in Section 3.    

4.     The second sentence in subsection 4.3 should be rewritten or use ‘than’ in place of ‘as well as’ in that sentence.

5.     Replace ‘Our primary goal of this paper is to’ with ‘Our primary goal is to’ at line 490.

 

 

 

Author Response

We express our gratitude to the reviewer for bringing these issues to our attention. We have incorporated all comments in the revised manuscript and we address each comment point-by-point below.

 

“The abbreviated terms, e.g., ‘EDA’, ‘HRV’ must be expanded on their first use.”

 We have revised abbreviations that were not expanded in their first use.

“In Section 3, data samples used for training the proposed model should be illustrated for its better understanding.”

We've provided a detailed explanation of the data segmentation process for training in Section 3.3.

“Author should mention the name of the learning algorithm used for supervised learning of the proposed technique and also specify the reason for adding one new FC layer in the fine-tuned pretrained model based on self-supervised in Section 3. ”

We have added the name of the learning algorithm for supervised training, 1D CNN,  in Section 3.4. We have included why FC layers are important for finetuning in the revised manuscript in Section 3.4

“The second sentence in subsection 4.3 should be rewritten or use ‘than’ in place of ‘as well as’ in that sentence.”

We have updated our revised manuscript as per the review.

“Replace ‘Our primary goal of this paper is to’ with ‘Our primary goal is to’ at line 490”

We have updated our revised manuscript as per the review.