Developing a Machine-Learning-Based Automatic Incident Detection System for Traffic Safety: Promises and Limitations

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper has developed a freeway traffic incident detection system using artifical neural networks. This is a very valuable research direction. However, several problems in the paper need to be solved to improve the quality of this research.

1. Literature reviews are too lengthy. Some of the content belongs to the explanation of basic knowledge, which is not necessary.

2. Although the shortcomings of different approaches are illustrated in the literature review, the problem to be solved in this paper is not clear. Is this study intended to address all these deficiencies?

3. For the study area selection, there is no detailed instructions. For example, how many junctions, merge areas, diverage areas in the sudy area? Authors should give a picture to exhibit the study area.

4. In data generation, authors only illustrate the limitations when using real data. However, using real data has many advantages. Therefore, I am not quite agree with the reasons given by authors in this part. Moreover, since the movement of vehicles in simulation software is simplified, it also will impact the utility of model. The calibration of the simulation model by the author is not seen in the paper, and whether it can truly reproduce the real traffic operation is uncertain.

5. Since the model is not validated by real data, Its effectiveness is questionable. It is suggested that authors add some real data to further verify the validity of the model.

6. The model-building part is too simple. Many of the modeling details are left unexplained. For example, why choose this structure? how to construct the model to solve the shortcomings mentioned in the literature review?

7. in 4.4, the maxium spacing is only 1.5km, which is not quite practical.

8. For the effect of incident location, to test the effect of incident happened in basic segment, junction, merge segment or diverage segment may be more valuable.

9. In page 16, line 719-721, what does that mean?

Comments on the Quality of English Language

Minor editing of English language required.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

1. The manuscript used of simulated data generated by VISSIM software to train and test the ANN model. Please provide further details regarding the simulation data generation process, including the selection of simulation parameters and how the simulated data ensures representation of real-world traffic scenarios.

2. The paper reported the model's performance on the test set but does not address the model's performance under other regions or traffic conditions. The authors are encouraged to discuss the model's generalizability and consider testing the model on additional datasets.

3. The paper mentioned a comparison of the ANN model with several other AID systems but does not provide an in-depth comparative analysis. Please provide a more detailed comparison, including the model's sensitivity, specificity, and performance under various traffic conditions.

4. The paper discussed the development and testing of the model but does not address the feasibility of deploying the model within actual traffic management systems. Please discuss potential challenges associated with model deployment, including hardware requirements, data transmission, and processing time.

5. As an ANN-based model, it is often considered a "black box." Please explore the interpretability of the model and how to explain the model's decision-making process to end-users, such as traffic management personnel.

6. Discuss directions for future work based on the limitations and findings of the current study, e.g., potential improvements, new research directions, or integration of new technologies.

Comments on the Quality of English Language

Need to be improved.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

The article introduces a promising idea for predicting traffic conditions and preventing accidents using a machine learning-based Automatic Incident Detection (AID) system. While the system shows impressive accuracy during testing, with a high Detection Rate and low False Alarm Rate, it faces concerns similar to those in weather forecasting. The accuracy of the system might be undermined by the numerous sensitive and unpredictable variables present in real-world traffic conditions. Just as weather forecasts can be affected by unexpected changes, the system’s performance could be compromised by the inherent complexity and variability of traffic, potentially limiting its effectiveness in consistently predicting and preventing incidents.

Unpredictable actions by drivers, such as sudden braking, lane changes, or distracted driving, introduce variability that the system might struggle to account for. Alcohol and drug effects can produce unpredictable patterns as well.

Other additional variables — oversize vehicles, space between vehicles relative to their speed, and the maximum and minimum speeds of vehicles in the same lane—can significantly impact the accuracy and reliability of an Automatic Incident Detection (AID) system.

Oversize vehicles, such as trucks or buses, typically occupy more space and move differently compared to regular vehicles. They may cause slower speeds, require more space for lane changes, and potentially block sensors or cameras that rely on line-of-sight, leading to data blind spots. The presence of oversize vehicles might cause the system to misinterpret traffic patterns, mistaking the slower or disrupted flow around these vehicles as an incident, or conversely, missing an actual incident hidden behind an oversize vehicle.

The space between vehicles, especially at varying speeds, is crucial for safety. At higher speeds, vehicles require more distance to stop safely. If the space is insufficient, it could lead to sudden braking or collisions, which the system needs to detect quickly. Variations in the space between vehicles can create complex patterns that the AID system must analyze. If the system is not well-calibrated to recognize these variations, it may struggle to differentiate between normal traffic conditions and potential incidents, especially in situations where vehicles are traveling at inconsistent speeds.

Significant differences in speed among vehicles in the same lane can lead to dangerous situations, such as abrupt braking or sudden lane changes. The system must accurately detect these discrepancies to predict potential incidents. Large speed variations might confuse the system, leading to false positives (detecting an incident when there isn’t one) or false negatives (failing to detect an actual incident). For example, if a vehicle is moving much slower than others, the system might flag it as an incident, even if it’s just due to congestion or a heavy load.

Incorporating these variables into the AID system's analysis could improve its ability to accurately assess and respond to real-time traffic conditions. However, it also adds complexity to the system, requiring more sophisticated algorithms and more detailed data collection to ensure accurate detection and minimize errors.

Despite the numerous improvements and additional variables that must be considered, the article presents a well-structured and consistent approach to advancing Automatic Incident Detection (AID) systems. It represents a significant step forward in developing more accurate systems capable of predicting and controlling traffic accidents.

The research’s comprehensive analysis and use of machine learning techniques demonstrate a clear progression in addressing the complexities of real-world traffic conditions. By incorporating a wide range of factors and employing advanced simulation tools, the study lays a solid foundation for improving the reliability and effectiveness of AID systems.

While challenges remain, such as accounting for unpredictable variables and refining the system's responsiveness, the article shows that we are moving closer to developing AID systems that can more accurately detect incidents and enhance traffic safety. The consistent methodology and promising results make this research a valuable contribution to the field, paving the way for future advancements in traffic management and accident prevention.

It might be wise to reconsider using the word "robust" in the title, as it could be perceived as too assertive given the model's current limitations. While the study makes significant progress in advancing Automatic Incident Detection (AID) systems, the complexity and unpredictability of real-world traffic conditions suggest that some academic modesty is warranted. A more measured title would better reflect the system's potential without overstating its capabilities, acknowledging that while the research is promising, there is still room for further refinement and validation.

 

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The author has made a better revision of the paper according to the reviewer's opinions and made a detailed reply. The revised manuscript has met the reviewer's requirements and can be considered for publication

 

Comments on the Quality of English Language

 Minor editing of English language required.

Author Response

Comment 1 : "The author has made a better revision of the paper according to the reviewer's opinions and made a detailed reply. The revised manuscript has met the reviewer's requirements and can be considered for publication"
Reply:  The authors sincerely thank the reviewer for their positive feedback and acknowledgment of the revisions made to the manuscript. We are pleased that the revised manuscript has met the reviewer’s expectations and is considered for publication.
Comment 2 : " Minor editing of English language required."
Reply:  We would also appreciate it if the reviewer could kindly point out any specific areas that may require attention.

Thank you again for your valuable feedback.

Reviewer 2 Report

Comments and Suggestions for Authors

4. Sorry but can not find the context the authors mentioned in the manuscript from lines 838 to 849.

5. There is no formulas throughout the paper. Please add essential formulas to tell the readers the methods or the basic equations in the machine learning model used in this study.

6. Sorry but can not find the context the authors mentioned in section 5.3 of the manuscript.

Comments on the Quality of English Language

Fine.

Author Response

Comment 4 : "Sorry but can not find the context the authors mentioned in the manuscript from lines 838 to 849"
Reply: The authors apologize for the mistake. The correct reference is to lines 862 to 872 on page 19 of the manuscript, not lines 838 to 849 as previously mentioned.
Comment 5: " There is no formulas throughout the paper. Please add essential formulas to tell the readers the methods or the basic equations in the machine learning model used in this study."
Reply : The formulas describing the working of the Multi-Layer Feedforward Artificial Neural Network (MLFANN) have been added to section 3.5, "Development of the AID Model using Multi-Layer Feedforward Artificial Neural Network (MLFANN)," in the revised manuscript.
Comment 6: "  Sorry but can not find the context the authors mentioned in section 5.3 of the manuscript."
Reply: The authors apologize for the typo. The correct reference is to section 6.3, "Recommendations for Future Research," not section 5.3.