Sustainable Data Construction and CLS-DW Stacking for Traffic Flow Prediction in High-Altitude Plateau Regions

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

1. It is recommended to provide a framework diagram of the entire prediction model and a description of the overall calculation process.
2.The title of 4.2 is not correct, and all the paper is all about speed prediction. It is recommended to revise it.
3 I suggest to explain why the accuracy of curve prediction is lower than that of continuous curve prediction.
4. It is recommended to delete the flowchart of the basic algorithm, as shown in Figure 8-12.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

My comments:

1. The abstract should be starting: "This paper present a plateau traffic flow...." to be clear.

2. Line 22. "segments.Furthermore"

3. Line: 29. "outputs.The experimental"

4. The abstract should be rewritten. Please add important data, like %, etc....

5.  Line 51-119, is it possible to put in a table with advantages and disadvantages?

6. Line 244, please explain with more details the table 1.

7. Line 254 "days.Based"

8. Figure 1. Please add. " Cumulative percentage in %"

9. Figure 1-4. What do you mean "Counts". 

10. Why do you use segformer method and not unet?

11. Line 490-494. Informer and your model, is it pretty similar or not? why?

12- The conclusions should be written in passive voice.

13. In the conclusions, please add important data. 

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The article addresses the current issue of predicting vehicle speeds in the challenging mountainous conditions of the Tibetan Plateau, which has great significance for traffic safety and policymaking. The authors combine deep learning methods (Bi-LSTM, Informer, GRU) with dynamic weighting through constrained least squares to create a new ensemble model (CLS-DW-Stacking). The use of drones for data collection in remote mountainous areas is an innovative solution to the problem of limited access to official data. The results show that the proposed model outperforms individual models in accuracy (MSE, RMSE, MAE, R²). The findings can serve as a basis for formulating speed limit policies on mountain roads.

The article is presented in a solid scientific structure; however, for minor improvements, I provide the following comments for consideration:

(1) Although the deep learning-based model is accurate, its results are difficult to interpret, which may hinder its application in practice or policymaking. The authors could include interpretable methods, such as SHAP analyses or decision trees, to clarify which factors have the greatest impact on the results.

(2) The model works with a limited number of inputs and does not account for factors such as weather, road surface conditions, or seasonal influences, which can significantly affect vehicle speed. Please consider extending the model to include environmental variables and demonstrate their effect on prediction accuracy.

(3) Data were collected from only two locations in Tibet, which limits the generalizability of the conclusions to other mountainous areas. The authors could add new datasets from different regions or at least discuss the extent to which the results are applicable beyond the studied areas.

(4) The model does not show how results change when parameters are adjusted (e.g., training sample size or weights). This may reduce the credibility of the results. Please consider performing a “what-if” (sensitivity) analysis to demonstrate the robustness of the model and strengthen the reliability of the conclusions, or at least discuss this aspect.

(5) The text is at times overloaded with technical details and contains minor language shortcomings, which may reduce readability even for a professional audience. Consider simplifying the presentation of technical details (possibly moving part of them to appendices).

(6) The current graphs and tables provide basic information but do not allow for sufficiently intuitive understanding of the differences between models. Please consider creating clearer graphical representations, such as maps or comparative charts highlighting prediction errors.

Questions for the Authors:

1. Why were only two vehicle types (standard vehicle and large truck) included in the model? Did this not reduce the diversity of the dataset?

2. How would the model respond to changes in weather conditions (e.g., snow, rain, low visibility)? Do you plan to include such factors in the future?

3. Since the data were collected only during the day and under favorable weather, could this lead to systematic bias in the results?

4. In what practical way does your CLS-DW-Stacking model differ from traditional ensemble methods (apart from the weighting approach)?

5. Have you considered incorporating more interpretable models (e.g., decision trees, SHAP analyses) to explain which factors most strongly affect speed?

6. How do you plan to validate the transferability of your results to other regions or types of mountain roads?

 

Thank you!

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

Now, this paper can be accepted!

Author Response

Thank you very much to the reviewer for recognizing this article. We sincerely wish the reviewer gratifying achievements in future research.

Reviewer 3 Report

Comments and Suggestions for Authors

The authors have addressed all comments and responded to the questions appropriately. The article can be published in its current form.

Best regards

Author Response

Thank you very much to the reviewer for recognizing this article. We sincerely wish the reviewer gratifying achievements in future research.