Applying Machine Learning Methods and Models to Explore the Structure of Traffic Accident Data

Round 1

Reviewer 1 Report

The topic of the article is interesting and presents results that deserve publication. There are, however, some issues that can be improved, such as:
- The conclusion section can be extended, by presenting several subsections: (1) contributions to the theory; (2) contributions to practice; (3) limitations; (4) suggestions for future research.
- The article presents the objectives (Introduction section) but, being a case study, it may make sense to present a research question.
- The literature review lacks some depth. I know the title refers to a "Brief Literature Review", but is there any reason to be brief?
Apart from these issues, it seems to me that the article is fine and has potential.

Author Response

Dear colleague,

thank you very much for your kind reply on our study.

1. The conclusion section can be extended, by presenting several subsections: (1) contributions to the theory; (2) contributions to practice; (3) limitations; (4) suggestions for future research.

We have tried to make our contribution to the theory and practice as well as limitations and further work more clear in the conclusion.

1. The article presents the objectives (Introduction section) but, being a case study, it may make sense to present a research question.

We also added a research question to the introduction part of the paper.

1. The literature review lacks some depth. I know the title refers to a "Brief Literature Review", but is there any reason to be brief?

We have changed the structure of the paper a bit and now this section is called "Related works".

We hope that our changes are acceptable for you.

Reviewer 2 Report

The paper evaluates an algorithm based on clustering techniques.
However, state-of-art does not warn about exploring these methods or instability due to their conjugate solutions. 
Methods such as K-means, Kmedoids, Mean Shift have been evaluated several times without being considered for real applications where centroid accuracy is required. A model error implies a different centroid location.
It is suggested that the authors evaluate state of the art in this regard.
The metrics are incipient, and an exhaustive evaluation is not performed.
It is recommended to review this type of method presented for other problems and what has been their result in georeferenced scenarios.

https://doi.org/10.1007/s11036-018-1002-6
https://doi.org/10.1007/978-3-319-73317-3_41
10.1109/ICMIC.2017.8321618
10.1109/ICC.2016.7511441
10.1109/ICCCN.2017.8038499

Authors are advised that references should be from recognized libraries Sciencedirect; MDPI; Wiley; IEEE Xplore [Transactions; Journal; Magazine]; Springer; Taylor & Francis. 
Authors are encouraged to perform a bibliometric analysis using VosViewer (Web of Science; Scopus) and identify authors and countries about clustering.

It is suggested to change the structure of the paper to 1.- Introduction; 2.- Related Works; 3.- Problem Formulation; 4.- Analysis of Results; 5.- Conclusions; 6.- References.

It is recommended to use a word processor such as Overleaf and include the metrics in PDF format. It is recommended to have a conceptual graph in section 2 that provides an overview of the problem and the proposed solution.

In the summary table of state of the art, the contribution and innovation of this work should be included.

It is suggested to find other metrics that show the innovation of this algorithm compared to other works presented.

Author Response

Dear colleague,

thank you for your feedback.

• However, state-of-art does not warn about exploring these methods or instability due to their conjugate solutions. 
  Methods such as K-means, Kmedoids, Mean Shift have been evaluated several times without being considered for real applications where centroid accuracy is required. A model error implies a different centroid location.
  It is suggested that the authors evaluate state of the art in this regard.

We add pairwise stability procedure to proove the stability of obtained results.

• Authors are advised that references should be from recognized libraries Sciencedirect; MDPI; Wiley; IEEE Xplore [Transactions; Journal; Magazine]; Springer; Taylor & Francis. 
  Authors are encouraged to perform a bibliometric analysis using VosViewer (Web of Science; Scopus) and identify authors and countries about clustering.

We have tried to use as many respectable sources as it was possible.

• It is suggested to change the structure of the paper to 1.- Introduction; 2.- Related Works; 3.- Problem Formulation; 4.- Analysis of Results; 5.- Conclusions; 6.- References.

The structure of the paper was changes according to your suggestion.

• It is recommended to use a word processor such as Overleaf and include the metrics in PDF format. It is recommended to have a conceptual graph in section 2 that provides an overview of the problem and the proposed solution.

There was introduces the conceptual scheme explaining problems and ways to solve them.

• In the summary table of state of the art, the contribution and innovation of this work should be included.

We have extended conclusion part with novelty issues.

 

 

Reviewer 3 Report

This paper introduced Machine Learning techniques can significantly increase the effectiveness of social marketing campaigns taking into account constructed typical “portraits” of the drivers violated road traffic rules within found target groups. In reviewer's opinion, the manuscript requires revision, in order to address the following concerns,

1. The novelty of the paper should be addressed in the Introduction section clear and simply.
2. There are many types of machine learning models, why did the authors choose LR, RR, DT, and NN? The author should elaborate on the architecture and model parameters of each model when applying in the data.
3. The authors should further enlarge the Introduction with current work of machine learning techniques to improve the research background, for example: Effective IoT-based Deep Learning Platform for Online Fault Diagnosis of Power Transformers Against Cyberattack and Data Uncertainties; Wind Farm Fault Detection by Monitoring Wind Speed in the Wake Region; Fusion of vibration and current signatures for the fault diagnosis of induction machines; Experimental setup for online fault diagnosis of induction machines via promising IoT and machine learning: Towards industry 4.0 empowerment.
4. The description of the dataset is not clear. For example, how many features were used?
5. The reviewer recommends a comprehensive English review. There are some minor issues.
6. All equations should be numbered.
7. In Figure 2, how to choose the optimal k based on Elbow approach? It could be 2, 6,7, or 8? Please explain more about that. In addition, the right side of fig. 2 is not described in detail. Did the authors verify the results with Silhouette Coefficient ‘s’?
8. Cross-validation is just a resampling procedure that is used to evaluate machine learning models on a limited data sample, it is not suitable to use to compare between models instate of classification accuracy, ROC curve of each model.
9. Is there any shortcoming of the developed framework? The authors are suggested to make some discussions.
10. The Conclusion needs to be revised to elaborate the summary of the study.

Author Response

Dear colleague,

thank you for the feddback on our study.

 

• The novelty of the paper should be addressed in the Introduction section clear and simply

We have tried to add formulation of study's novelty in the introduction part.

• There are many types of machine learning models, why did the authors choose LR, RR, DT, and NN? The author should elaborate on the architecture and model parameters of each model when applying in the data.

We have analyzed latest researches related to applying ML techniques in the transportation and found that these classical models are commonly used. We also have an expirience of solving similar problems with such models. We also added information on structure and optimally found (usind grid search approach) parameters for all presented models.

• The authors should further enlarge the Introduction with current work of machine learning techniques to improve the research background, for example: Effective IoT-based Deep Learning Platform for Online Fault Diagnosis of Power Transformers Against Cyberattack and Data Uncertainties; Wind Farm Fault Detection by Monitoring Wind Speed in the Wake Region; Fusion of vibration and current signatures for the fault diagnosis of induction machines; Experimental setup for online fault diagnosis of induction machines via promising IoT and machine learning: Towards industry 4.0 empowerment.

We add additional examples of using effective ML techniques in solving similar problems.

• The description of the dataset is not clear. For example, how many features were used?

We have tried to give more detailed description of used data set.

• The reviewer recommends a comprehensive English review. There are some minor issues.

Thank you. Manuscript was checked by professional translator.

• All equations should be numbered.

Done.

• In Figure 2, how to choose the optimal k based on Elbow approach? It could be 2, 6,7, or 8? Please explain more about that. In addition, the right side of fig. 2 is not described in detail. Did the authors verify the results with Silhouette Coefficient ‘s’?

We added explanation on optimal number of clusters choosing.

• Is there any shortcoming of the developed framework? The authors are suggested to make some discussions.

We have tried to discuss obtained results deeply.

• The Conclusion needs to be revised to elaborate the summary of the study.

Done.

We hope that our changes are acceptable for you.

Round 2

Reviewer 2 Report

Figure 5 could be significantly improved using Matlab. To save the image in PDF format, you can use the command print -dpdf -r800 figure5; then, you can include the figure in your Overleaf project.

On the other hand, the authors are encouraged to innovate concerning the following previously published works on the clustering process in geo-referenced scenarios.

- https://doi.org/10.1007/s11036-018-1002-6
- https://link.springer.com/chapter/10.1007/978-3-319-73317-3_41
- https://ieeexplore.ieee.org/document/7909154
- https://ieeexplore.ieee.org/document/8321618
- https://ieeexplore.ieee.org/document/7511441
- https://ieeexplore.ieee.org/document/8038499
- https://doi.org/10.3390/en13010097

Author Response

Thank you for the feedback and valuable comments.

1. Figure 5 could be significantly improved using Matlab. To save the image in PDF format, you can use the command print -dpdf -r800 figure5; then, you can include the figure in your Overleaf project.

Figures were improved and saved in high resolution from Python.

1. On the other hand, the authors are encouraged to innovate concerning the following previously published works on the clustering process in geo-referenced scenarios.

The authors have tried to add information on proposed studies and showed that these researches belong to other direction of applying k-means technique. 

Reviewer 3 Report

The paper has improved with the revisions. The reviewer has no more concerns so far.

Author Response

Thanks a lot for your feedback!