A Deep Learning Model Development to Predict Safety Accidents for Sustainable Construction: A Case Study of Fall Accidents in South Korea

Round 1

Reviewer 1 Report

A deep learning model development to predict safety accidents for sustainable construction is presented in the research. It is a case study of fall accidents in South Korea. The article is well written, however some minor corrections are required to improve the quality of the publication. 

Line 44 to 87 text is bold (please check)

The heading of Section 2 should be "Literature Review"

In Section 3 the heading should be "Research Objectives" because more than one objectives are mentioned here.

The sections 5 could be merged with Section 4 because data collection comes under the research methodology. It would be good if the authors add a figure here

Table 1: In first column the logically the input should come first then output (along with other parameters e.g., variable, type and description)

Same observation is for table 2.

Please check "436" in Table 3 column 1

The font used in the Section 6 is bold (Lines 312 to 345) please check

 

 

Author Response

We answered all the reviewer's comments. We attach the revised file.

Author Response File: Author Response.pdf

Reviewer 2 Report

1. The reason for using  ReLU activation function in the hidden layer and softmax function in the output layer.
2. Why only 3 hidden layers are considered? 
3. In data preprocessing only z-score normalization is applied but what about other redundant or null or duplicate data?
4.  Which optimizer is used?
5. DNN is compared with MRA but DNN must be compared with other Deep Learning models.
6. Try to include variables for Employee age and gender also.
7. Safety measures as per standard are provided or not.
8. Carry out the experiment for an odd number of nodes in hidden layers.
9. 1,766 cases of actual accidents were collected by  KOSHA over the 10-year period from 2010 to 2019 but the type of accidents- minor, major, no. of peoples injured in one accident must be considered.

Author Response

We answered all the reviewer's comments. We attach the revised file.

Author Response File: Author Response.pdf

Reviewer 3 Report

1. Innovation points are not condensed enough. Each point should include questions, methods, results and conclusions.For example, “for the xx problem, the xx method is used. We verify the effect through the xx experiment, and the conclusion is xxx.”
2. Whether the author can increase the theoreticality of the article by formula.
3. Figure 1 does not reflect the theme of the article very well. It is recommended that the author revise it, and use more graphics and colors, not just words!
4. When conducting experimental analysis, the author should not simply state the experimental results, but also explain the reasons for the results. In this way, the advantages of the method can be better reflected.
5. The conclusion section is too cumbersome to summarize. In addition, the conclusion part lacks prospects for future work.
6. Authors should carefully check the statements in the article.

Author Response

We answered all the reviewer's comments. We attach the revised file.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

After english editing; can be processed

Author Response

We checked the manuscript overall.

Author Response File: Author Response.docx

Reviewer 3 Report

The work of this paper is practical and logical. However, it still exists the following problems:
1.Your reply was perfunctory.
2.You didn't make reasonable changes according to the review comments, and didn't make careful changes in many places, such as the comments on pictures in review comment 3.

 

Author Response

We revised the paper reflecting the reviewer's comments.

Author Response File: Author Response.docx

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

1. Regarding the data set of the model, the author only cited the conclusions of other papers and selected eight variables. For different scenarios, the experimental settings are different, so the author's direct conclusions are not convincing. To ensure the scientificity and rigor of the experiment, could the author consider selecting variables through some dimensionality reduction operations of component analysis? Please provide theoretical or experimental results support.
2. My main concern is the Originality of the paper. The MAE error and RMSE error used by the authors are widely used by predictive and classification models in ML. However, none of them are original as they are already well-developed in the community. And, only the ordinary DNN network is used to adjust the number of nodes to complete the experiment. In Section 6, the process of the experiment should have been emphasized, but the author spent a lot of space describing the common sense content in the field of ML. There is too much unnecessary common sense content. For the specific experimental scene of safety accident prediction, how to carry out the specific and creative design of this scene is very few.
3. The author only uses 1766 pieces of data to train the model. How to deal with factors such as noise data? In the future, in the face of new application scenarios, new variables may be introduced. At this time, how does the safety accident prediction model proposed by the author respond?
4. There is no comparison with existing methods, and the effectiveness is not convincing.
5. The model and the experimental process lack graphic descriptions, please add the corresponding intuitive structure diagrams and other supplementary materials
6. The experiment process lacks explanation, scientificity and rigor are not persuasive.
7. There are some unnecessary content of common sense in this work, such as Figure 1
8. There are some grammar errors, such as ‘have’ on line 26.

Author Response

We answered all the reviewer's comments. We attach the file.

Author Response File: Author Response.pdf

Reviewer 2 Report

A case study of fall accidents in South Korea is presented in the research. The purpose of the study is a deep learning model development to predict safety accidents for sustainable construction. The results of the study provide a guidelines for the introduction of deep learning technology in construction safety management.

The structure of the article should be revised to improve the quality of the paper. Some important information is missing while the presented information is not properly organized. Following is the guidelines to improve the quality of the paper.

1. Please provide the structure of the article at the end of the introduction section.
2. The section 2 and 3 also have sub-sections, which are starting just after the main heading. It is suggested to add few lines to explain the sub-sections at the beginning to create interest to the reader.
3. Adjust the text inside table 1, use suitable spacing (in current form it is not properly readable)
4. What did you achieve? There is no results section!
5. What is the purpose of providing Figure 2 in discussion section?
6. Conclusion section is too big and should be revised (decrease the word count) to leave a strong final impression to the reader. Also provide some future directions to extend the work.

Author Response

We answered all the reviewer's comments. We attach the file.

Author Response File: Author Response.pdf

Reviewer 3 Report

1. Explain in detail the reason for using  ReLU activation function in the hidden layer and softmax function in the output layer.
2. Justify the use of only 3 hidden layers are considered? 
3. In data preprocessing only z-score normalization is applied but what about other redundant or null or duplicate data?
4.  Which optimizer is used?
5. DNN is compared with MRA but DNN must be compared with other Deep Learning models.
6. Try to include variables for Employee age and gender also.
7. Safety measures as per standard are provided or not.
8. Carry out the experiment for an odd number of nodes in hidden layers.
9. 1,766 cases of actual accidents were collected by  KOSHA over the 10-year period from 2010 to 2019 but the type of accidents- minor, major, no. of peoples injured in one accident must be considered.

Author Response

We answered all the reviewer's comments. We attach the file.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The manuscript has been revised and recommended for publication in present form.

Reviewer 3 Report

Authors have incorporated suggestions given by me