The Urban Intersection Accident Detection Method Based on the GAN-XGBoost and Shapley Additive Explanations Hybrid Model

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

A method to establish the reasons of traffic accidents in urban intersections is suggested. The paper subject is suitable for publication in Sustainability since the most noticeable features of the model are described together with the site where the model is applied. However, some minor changes should be considered prior to its acceptance.

The name of the city should be included. A noticeable doubt arises about the data representativity, i.e., readers could wonder if this model could be applied to different cities with different size or where driving habits may be quite different.

Another issue is focused about conditions linked to the traffic accidents. For instance, readers should know if the model was tested under adverse weather conditions.

Minor remarks.

The paper structure does to respond to that for a typical paper where the Material and Methods Section is placed between the introduction and the results sections. In the Material and Methods section all the data and equations should be placed. Consequently, the case study description belongs to Material and Methods section, whereas 4.2 section belongs to results.

Moreover, some references should be included in the Results and discussion section to establish the contrast between this research and previous analyses.

L. 184. Revise “This We”.

L. 212. Revise “…accident, In…”.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The paper addresses the impact of traffic accidents on congestion using xgboost. The paper sounds interesting but has a few open issues to be tackled as follows.


Methdolology:

the authors did not use a sufficient statistics for the congestion evaluation. Traffic speed and weather conditions are one issue.
Communication while driving (cellphone, talking to neighbor, etc.) and alcohol are other issues that strongly influence accidents. Did the authors make sure that the driver
was sober and solely focused on the driving during the time of the experiment. If not, the inferred result on traffic speed as reason is incomplete.
To be clarified.


Introduction:

Machine learning techniques in general play an important role in road accident forecasting (DOI: 10.1016/j.trip.2023.100814). An overview of machine learning techniques for accident modeling should be added to the list of literature.

Moreover, data congestion (caused by rate and speed) is strongly related to vehicle congestion. For example, DOI 10.1109/JIOT.2022.3142324(IEEE Internet of Things), the authors followed a stochatic approach using data-flow graphs to model data congestion. Stochastic data-flow graphs for congestion modelling should be added, as well.



Results/Discussion:

The authors came to conclusion that XGBoost outperform RF while the authors in DOI:  10.1016/j.trip.2023.100814 (metioned above) came to the opposite conclusion.  How can that be explained?

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The authors tackled all open issues. Good work!