A Review of Traffic Flow Prediction Methods in Intelligent Transportation System Construction
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe study is dedicated to analyzing methods for traffic flow prediction in terms of their effectiveness in Intelligent Transportation Systems (ITS).
The authors consider three main groups of methods:
Traditional statistical models
Machine learning models
Deep learning models
The topic of the study is relevant. The authors summarize a wide range of methods and their specific applications for traffic flow prediction, which can serve as a solid basis for identifying vulnerabilities in this field.
However, since the authors state that Intelligent Transportation Systems operate in real time, I believe it would be appropriate to include adaptive real-time prediction models in the discussion. This addition could improve the performance of ITS when working with streaming input data.
When comparing different groups of methods, the authors identify advantages and disadvantages based on several datasets. However, I believe it would also be useful to strengthen the analysis of model performance and computational complexity in real-time settings. This could involve specifying certain parameters such as training time, prediction time, and memory usage. The limitations of computational resources required by specific models may become a key factor in determining their applicability in real-time systems.
The conclusions made by the authors are consistent with the results of the study. However, to increase the practical value of the research, it would be helpful to provide conclusions about the applicability of specific models for various traffic flow prediction tasks. This may include considerations based on the size of training data, technical limitations of computing devices, prediction time steps, and other relevant factors.
Author Response
Dear reviewer,
We sincerely appreciate the time and effort that you have devoted to evaluating our manuscript titled “A Review of Traffic Flow Prediction Methods in Intelligent Transportation System Construction” (ID: applsci-3547839). We are grateful for the insightful comments, which have significantly contributed to improving the quality of our work.
In response to the comments, we have carefully revised our manuscript.
A detailed point-by-point response to reviewercomment is provided in the attached response document. We have carefully addressed all concerns raised and incorporated the necessary changes accordingly.
We hope that the revised manuscript meets the journal’s standards and expectations. We appreciate your consideration and look forward to your favorable response.
Best regards,
Runpeng Liu,liuy31814@gmail.com
Corresponding author:
Name:Seong-Yoon Shin
E-mail:s3397220@kunsan.ac.kr
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThis paper is entitled “A Review of Traffic Flow Prediction Methods in Intelligent Transportation System Construction” In this case, the idea and results of the paper are interesting but the following comments can be utilized to improve this paper in future.
Abstract
- Remove “1. Introduction” at the end of abstract
- Sentence structure and conciseness: Some sentences are overly complex or repetitive and can be streamlined for better readability. Example: "Accurate traffic flow forecasting can significantly support traffic management and decision-making."
- Improve the transition between ideas. Issue: The shift from classification to the role of deep learning feels abrupt.
- Clarify "pivotal role of deep learning models." Issue: The abstract states that "deep learning models play a pivotal role in improving prediction accuracy, particularly in short-term forecasting," but does not specify how.
- Avoid vague phrases like "new perspectives." Example: "Aiming to provide new perspectives and enhance the development of ITS technologies."
- Expand on statistical and machine learning methods. The abstract focuses heavily on deep learning but does not sufficiently explain the relevance of statistical and machine learning-based approaches.
- Lack of real-world applications. The abstract mentions practical case studies but does not briefly summarize key findings.
Introduction and literature
- Some sentences are lengthy and difficult to read. Check it for all part of the manuscript. Example: "The prediction models LSTM and GRU based on deep learning have greatly improved the prediction accuracy. The prediction model based on hybrid neural network ASTGCN has the best overall prediction accuracy because it takes into account the temporal and spatial correlation."
- Improve transitions between sections. The transition from Table 2 (model performance comparison) to Table 3 (advantages/disadvantages) is abrupt.
- Lack of explanation for why specific models perform better. Example: The section states: "Deep learning has more accurate prediction results, faster calculation speed, and can meet the requirements of high-precision real-time prediction, with obvious advantages." However, it does not explain why deep learning models outperform traditional methods.
- No discussion of computational cost or scalability. The section does not mention the resource requirements for different models.
- Tables should have more detailed descriptions. Table 2 lacks explanations for the evaluation metrics (MAE, RMSE).
- Table 3 should categorize deep learning models more clearly.Example: Instead of listing CNN, RNN, LSTM, and GRU separately, group them into:
- Recurrent Neural Networks (RNN-based models): RNN, LSTM, GRU
- Graph-based Neural Networks: GCN, ASTGCN
- Transformer-based Models: Transformer
- No mention of hybrid models or emerging trends. The section briefly mentions hybrid models (e.g., ASTGCN) but does not discuss other approaches such as:
- Meta-learning (models that adapt to new traffic environments quickly).
- Federated learning (models that learn from distributed data without centralization).
- Explainable AI (XAI) (interpretable models for traffic flow prediction).
Conclusion
- Sentence structure needs refinement for better readability. Example: "After discussing the principles, advantages, limitations, and applications of these methods in intelligent transportation systems, we found that with the rapid development of big data and artificial intelligence technologies, deep learning models have shown obvious advantages in traffic flow prediction, especially when dealing with large-scale, nonlinear, and high-dimensional data, their prediction accuracy and generalization ability are significantly better than traditional methods."
- Lack of specific examples or references. Example: The text states that deep learning performs better with "large-scale, nonlinear, and high-dimensional data" but does not specify any models or real-world applications.
- Vague statements in the discussion of dataset limitations. Example: "The data sets used in the current research are mainly from urban road traffic flow data. Therefore, the number of abnormal data samples in the training set is smaller than that of normal data, and the applicability to other types of areas (such as rural or suburban areas) still needs to be further verified."
- Repetitive phrasing and lack of clear sectioning. Example: The discussion of long-term vs. short-term traffic flow prediction is important but feels repetitive and lacks transition.
- Lack of emerging trends and novel solutions. Example: The text suggests merging different data types (e.g., weather, population, images) but does not mention new techniques such as federated learning, explainable AI (XAI), or transfer learning.
Final decision: This manuscript has interesting objectives, organization, and results. it needs Minor correction, then it is suitable to publish.
Author Response
Dear reviewer,
We sincerely appreciate the time and effort that you have devoted to evaluating our manuscript titled “A Review of Traffic Flow Prediction Methods in Intelligent Transportation System Construction” (ID: applsci-3547839). We are grateful for the insightful comments, which have significantly contributed to improving the quality of our work.
In response to the comments, we have carefully revised our manuscript.
A detailed point-by-point response to reviewercomment is provided in the attached response document. We have carefully addressed all concerns raised and incorporated the necessary changes accordingly.
We hope that the revised manuscript meets the journal’s standards and expectations. We appreciate your consideration and look forward to your favorable response.
Best regards,
Runpeng Liu,liuy31814@gmail.com
Corresponding author:
Name:Seong-Yoon Shin
E-mail:s3397220@kunsan.ac.kr
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe presented paper is dedicated to methods of traffic flow prediction in the context of intelligent transportation system development, which is an extremely relevant topic given the rapid development of the smart cities concept. The article covers various approaches to traffic flow prediction, classifying them into three main categories: statistical methods, machine learning methods, and deep learning-based approaches. The authors attempt to cover a wide spectrum of methodologies, analyzing their principles, advantages, limitations, and specific applications in intelligent transportation systems.
Shortcomings of the Paper
Despite the broad coverage of various prediction methods, the article lacks a clear formalization of the model. The authors do not provide a mathematical formulation of the traffic flow prediction problem, which could serve as a unified basis for comparing different approaches. Without such formalization, it is difficult to assess how different methods correspond to the specific requirements of the traffic flow prediction task.
A significant portion of the article is devoted to commonly known information about model structures and evaluation metrics. Instead, it would have been more appropriate to focus on calculation details, particularly the input parameters of methods, ways of obtaining them, and how this solves problems using known frameworks and software tools. Such an approach would have given the work greater practical value for researchers and engineers working in the field of intelligent transportation systems.
The work lacks analysis of methods that use commonly available resources such as Google, Waze, Open Street Map. This is a significant drawback, as these platforms provide vast amounts of data on traffic flows and have their own prediction algorithms that are widely applied in practice. Including analysis of these resources would have enriched the work and made it more relevant to the current state of the field.
In Table 4, the authors provide a comparison of the accuracy of different models, but the methodology for obtaining these data remains unclear. It is not specified which datasets were used for testing, which model parameters were chosen, and how exactly the accuracy was evaluated. Without this information, it is difficult to assess the reliability of the presented results and their applicability in real conditions.
The authors describe various configurations of machine learning and artificial intelligence models that are traditionally used for different classes of tasks, but do not explain how these different approaches are adapted to solve the specific task of traffic flow prediction. The article lacks a structural plan that would explain how to combine all these methods for effective traffic flow prediction. This creates the impression that the authors are simply listing existing methods without providing recommendations for their integration and practical application.
Author Response
Dear reviewer,
We sincerely appreciate the time and effort that you have devoted to evaluating our manuscript titled “A Review of Traffic Flow Prediction Methods in Intelligent Transportation System Construction” (ID: applsci-3547839). We are grateful for the insightful comments, which have significantly contributed to improving the quality of our work.
In response to the comments, we have carefully revised our manuscript.
A detailed point-by-point response to reviewercomment is provided in the attached response document. We have carefully addressed all concerns raised and incorporated the necessary changes accordingly.
We hope that the revised manuscript meets the journal’s standards and expectations. We appreciate your consideration and look forward to your favorable response.
Best regards,
Runpeng Liu,liuy31814@gmail.com
Corresponding author:
Name:Seong-Yoon Shin
E-mail:s3397220@kunsan.ac.kr
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for AuthorsThe authors review the problem of road traffic flow predictions in complex
environments which are an essential ingredient in ITS. They discuss three main approaches, i.e. statistical, machine learning and deep learning ones. The method are critically presented and their different performances and disadvantages are discussed. As such, the review appears at a first glance useful to readers looking for a compact and detailed summary of these models. However, I think a comparison between real data and predictions should be included. In the following I include my suggestions which may improve the setup of their work.
(1) The number of data used, the variable n in Eqs.(11,12,13) should be specify. It would be useful to check how the results depend on n, to provide a criterion for the stability of the conclusions.
(2) Disadvantages: In Table 3 it is often mentioned that some models cannot deal accuratley with non-stationary situations. This is because they assumed a stationary process, such as in the case of ARIMA. A sentence should be added stressing this feature, and to which models statitionarity is assumed.
Question: Are Deep learning models able to deal with non-stationary situations?
(3) Unfortunately, there is a lack of information which prevent the reader a more quantitative understanding of the problem. I recommend to add examples of how the (say the best) methods perform in particular cases. It should be clear what it is predicted and how the result is obtained. Provide a figure (or figures) with a typical measured traffic flow on top of which the predictions are shown. These additional results should be added to Section 6.
Author Response
Dear reviewer,
We sincerely appreciate the time and effort that you have devoted to evaluating our manuscript titled “A Review of Traffic Flow Prediction Methods in Intelligent Transportation System Construction” (ID: applsci-3547839). We are grateful for the insightful comments, which have significantly contributed to improving the quality of our work.
In response to the comments, we have carefully revised our manuscript.
A detailed point-by-point response to reviewercomment is provided in the attached response document. We have carefully addressed all concerns raised and incorporated the necessary changes accordingly.
We hope that the revised manuscript meets the journal’s standards and expectations. We appreciate your consideration and look forward to your favorable response.
Best regards,
Runpeng Liu,liuy31814@gmail.com
Corresponding author:
Name:Seong-Yoon Shin
E-mail:s3397220@kunsan.ac.kr
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThe authors have adequately addressed all of the comments, and the manuscript has been sufficiently improved. I recommend it for publication in Applied Sciences.
Reviewer 3 Report
Comments and Suggestions for AuthorsI am satisfied of author corrections
Paper can be published
I am satisfied of author corrections
Paper can be published
Reviewer 4 Report
Comments and Suggestions for AuthorsThe authors have considered all the questions raised and responded accordingly. After reading the revised manuscript I can recommend publication in Applied Sciences.