Prediction Modeling of Ground Subsidence Risk Based on Machine Learning Using the Attribute Information of Underground Utilities in Urban Areas in Korea

Round 1

Reviewer 1 Report

This is a very nice study of high importance. The manuscript assesses the subsidence risk of urban areas in Korea by a machine learning (ML) algorithm after comparing the performance of 3 different ML algorithms. Finally, the manuscript provides important subsidence risk information relevant for precaution action for different areas in Korea. Although, the study is well designed and the analysis is carefully performed, the manuscript needs extensive modifications prior it is considered for publication in Applied Science. One main problem is the language of the manuscript which at many points it is hard to understand. I would suggest that the authors carefully read the manuscript and improve its language. Additionally, some further comments are listed below:

-   Introduction lines 67-68: "...a study was conducted... in Korea through logistic regression analysis." Add here the corresponding citation.

- Introduction line 73: " ... that use various approaches to predicting risk...". Either predicting or to predict.

- Introduction. Can the subsidence risk also be assessed by finite element analysis? The authors should add a paragraph discussing this aspect and if any literature on the topic.

- 2.2 page 4, lines 135-137: "The risk levels were categorized.... as the risk levels were categorized further." This sentence is difficult to understand. Please explain better.

- 2.2. page 4 lines 141-143 and Table 1. It is difficult to understand the classification to risk levels 2 and 3. First for level 2 the authors state that the N of occurrences is 1, 1-2 or 1-3. Does Risk level 2 mean that N of ground subsidence occurrences = 1-3 or else what is the meaning of the 3 different cases (one, one to two or one to three)? Additionally, the authors should explain a little bit further when risk level 2 or 3 is assigned. For example, if the number of ground subsidence is 2 which risk level is assigned?

- Table 4: Does it mean that Corr is calculated on the input data of each factor and on the output data?

- 3.1-3.3 The authors should discuss a bit more the ML algorithms which is the core of the study. Moreover, did the authors use some prefixed computational platform, e.g. MatLab, R, python, for the ML analysis?

- It is not completely clear to me the purpose of pre-classifying the 24 datasets to 12 models. What would be the expectations of the results if the ML analysis performed on the entire dataset before it has been pre-classified. The authors should share some justification of their approach.

- Conclusions. The authors should not summarize the manuscript but rather clearly present their important findings. I would suggest that the conclusions cover the following directions (i) best ML performance model (ii) importance of influence parameters (iii) risk maps and (iv) suggested actions and future work. 

I suggest that the authors avoid the long sentences and try to use more comprehensive language.

Author Response

첨부 파일을 참조하십시오

Author Response File: Author Response.docx

Reviewer 2 Report

In this study, the authors proposed a prediction modeling of ground subsidence risk based on the attribute information of underground utilities in urban areas in Korea. Although the performance seems promising, some major points should be addressed as follows:

1. There must be external validation data to evaluate the model's performance.

2. How did the authors perform hyperparameter tuning of machine learning models?

3. The authors should compare the predictive performance to previously published works on the same problem/data.

4. Uncertainties of models should be reported.

5. It is unclear why the authors selected the current three machine learning models rather than others.

6. Model's performance can be improved using some advanced models, even deep learning.

7. Some results were listed without in-depth discussions.

8. Machine learning is well-known and has been used in previous studies i.e., PMID: 33848577, PMID: 28285094. Therefore, the authors are suggested to refer to more works in this description to attract a broader readership.

9. The authors should plot a heatmap for correlation analysis.

10. Overall, English writing should be improved.

11. Some figures/tables are unnecessary i.e., Fig. 2, Table 5, Fig. 3, etc.

12. The authors should plot ROC and PR curves. 

Overall, English writing should be improved.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

No further comments.

Reviewer 2 Report

My previous comments have been addressed well.

Minor editing for publication quality.