All reviews of published articles are made public. This includes manuscript files, peer review comments, author rebuttals and revised materials. Note: This was optional for articles submitted before 13 February 2023.
Peer reviewers are encouraged (but not required) to provide their names to the authors when submitting their peer review. If they agree to provide their name, then their personal profile page will reflect a public acknowledgment that they performed a review (even if the article is rejected). If the article is accepted, then reviewers who provided their name will be associated with the article itself.
Congratulations! All the suggestions have been addressed.
[# PeerJ Staff Note - this decision was reviewed and approved by Shawn Gomez, a PeerJ Section Editor covering this Section #]
in the reviewed manuscript, the authors clearly introduce addressed research problem as well as the main contributions.
My comments were addressed in the reviewed manuscript.
no comment
Please, revise the comments and suggestions of the reviewer 1.
For Table 1 : add the type of data and the objective of the prediction
How data are represented as input for the CNN
What are the inputs and outputs of the model. What are the predicted targets ?
Did the authors used cross validation instead of simple validation.
I suggest evaluating the model without the attention mechanism in order to assess its impact.
no comment
Dear Authors,
After careful evaluation by two expert reviewers, we have reached a decision. We appreciate your contribution, which presents a technically sound and potentially impactful model for traffic congestion prediction in smart cities. However, both reviewers have recommended that major revisions are necessary before your manuscript can be considered for publication.
Please find below a summary of the reviewers' main comments:
Reviewer 1 highlighted the need to:
• Improve the introduction, especially regarding the novelty of your approach, given the abundance of existing ML and DL models for traffic prediction.
• Justify the use of CNN in your hybrid model—specifically, whether spatial dependencies (as in image-like data) are present in your dataset.
• Explore different hyperparameters to strengthen the robustness of your findings.
Reviewer 2 provided a detailed assessment and pointed out the following areas for improvement:
• The literature review table (Table 1) conflates strengths and limitations and would benefit from clearer structure and critical analysis.
• The attention mechanism, although novel, requires a more detailed explanation (e.g., implementation details, alignment score computation); inclusion of diagrams or pseudocode is strongly encouraged.
• Dataset limitations, particularly the missing/incomplete data in Junction 4, should be addressed more rigorously (e.g., with imputation or sensitivity analyses).
• Missing figures referenced in the text (e.g., model architecture, prediction plots) should be included.
• Minor errors (e.g., typographical issues, inconsistent references, and the overly dense Table 3) should be corrected.
We believe that addressing these comments will significantly enhance the quality and clarity of your manuscript. Please ensure that your revised submission includes a point-by-point response to each of the reviewers' comments, outlining the changes made or justifying any aspects that were not modified.
We look forward to receiving your revised manuscript within the allocated revision period.
In the introduction, authors should improve the motivation, as many ML and DL models have been developed for traffic prediction. What is the main novelty of the article?
Used data and methodology used are well defined. One main question is why authors are using CNN. Are there any kind of spatial dependencies in the data (like images)?
The authors detail the results well. However, it is highly recommended to test different hyperparameters.
This paper proposes a hybrid deep learning model combining Bi-LSTM, CNN, and a custom attention mechanism to predict traffic congestion in smart cities. The model is evaluated on a dataset of 48,120 observations across four junctions, using RMSE, MSE, and MAE metrics. Results demonstrate superior performance over baseline models (GRU, LSTM, CNN, MLP), particularly in Junctions 1 and 2, though Junction 4 shows higher errors, attributed to data complexity. The work emphasizes preprocessing, feature engineering, and the novel integration of Bi-LSTM with attention.
The paper presents a technically sound and impactful contribution to traffic prediction. The model is evaluated on a dataset of 48,120 observations across four junctions, using RMSE, MSE, and MAE metrics. I would kindly ask the reviewer to focus on Junction 4, which shows higher errors, attributed to data complexity.
Technical Contribution: The hybrid architecture (CNN-BiLSTM-Attention) is well-motivated and addresses gaps in existing methods. The use of Bi-LSTM for bidirectional temporal dependencies and a custom attention layer adds novelty.
Empirical Validation: Extensive experiments across four junctions with clear metrics (RMSE, MSE, MAE) validate the model’s effectiveness. Results are statistically significant, especially in J1 (RMSE = 0.252) and J2 (RMSE = 0.561).
Practical Relevance: The work aligns with smart city initiatives and highlights real-world applicability, including preprocessing steps and dataset limitations
Clarity in Literature Review: Table 1 conflates advantages and limitations (e.g., Du et al. [9] lists "accuracy" as a limitation). A clearer distinction and critical analysis of prior work would strengthen this section.
Attention Mechanism Details: While the custom attention layer is mentioned, its implementation (e.g., weight initialization, alignment score computation) lacks depth. A diagram or pseudocode would enhance reproducibility.
Data Limitations: The dataset’s missing/incomplete data (notably in J4) is acknowledged but not rigorously addressed. Techniques like imputation or sensitivity analysis could be discussed.
Visualization: Figures referenced (e.g., architecture diagrams, prediction plots) are absent in the provided text, limiting assessment of visual results.
The paper presents a technically sound and impactful contribution to traffic prediction. However, revisions are needed to:
Clarify the literature review table and expand on the novelty of the attention mechanism.
Address dataset limitations (e.g., handling missing data in J4).
Include missing figures/visualizations to support results.
Correct minor inconsistencies and typos, e,g, Typographical errors (e.g., "traffic flow price" in Equation 2 should be "traffic flow rate").
Inconsistent reference formatting (e.g., mix of full names and initials).
Table 3 is overly dense; summarizing key epochs (e.g., final validation loss) would improve readability.
All text and materials provided via this peer-review history page are made available under a Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.