Machine Learning Traffic Flow Prediction Models for Smart and Sustainable Traffic Management
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis paper explores the use of multisource data to predict traffic pattern using machine learning technique for Melbourne’ eastern freeway. This paper presents Bi-directional LSTM model with the upstream and downstream detector dataset. It also investigate the role of special flow input interactions from upstream and downstream detectors.
The paper is well-structured, sophisticated, and contributes meaningfully to traffic flow prediction modelling.
- The study addresses an important urban challenge, short-term traffic flow prediction, and links it to sustainability goals such as reduced emissions, SDG alignment etc.
- The use of BiLSTM models and comparisons with LSTM, RNN, Elman, and DLBP models provide technical depth.
- The methodology of using multisource data was sound and clear.
- The paper effectively ties technical results to the United Nations Sustainable Development Goals.
- Showing accuracies across combinations of variables for each detector makes it interpretable.
Suggestions/Questions:
- The references in the main text (e.g., “(Bartlett et al., 2018)”) should follow MDPI formatting using square brackets.
- Figure 5, Figure 6 are referenced but not always cited in order. Please ensure they appear in sequence.
- Figure 7’s visual alignment with SDGs is discussed but not clearly shown.
- While the paper uses other model, it only shows BiLSTM results, The author should include comparative performance.
- The paper provides detailed BiLSTM settings such as 300 hidden units, learning rate = 0.005, but lacks justification for these choices or whether a tuning process (grid/random search) was applied.
- The explanation of results for each detector is verbose and repetitive. Consider summarizing the trends in less sentences.
- Missing diagram of LSTM model. Can refer this article:
https://www.mdpi.com/2412-3811/7/11/150
This paper is methodologically strong and well-aligned with sustainability objectives, making it highly relevant to the field of intelligent transportation systems and smart mobility. With some formatting cleanup, minor restructuring, and deeper comparisons, it would be a strong contribution to the literature.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe manuscript proposes a traffic flow prediction model based on Bidirectional Long Short-Term Memory Network (BiLSTM), which significantly improves the prediction accuracy through multi-source data (flow rate, speed, occupancy rate) and spatial interaction (data from upstream and downstream detectors). The research results have practical significance for intelligent transportation systems and the Sustainable Development Goals (SDGs). The following suggestions for modification and improvement are put forward:
1. When compared with other models, although the advantages of the BiLSTM model were mentioned, the analysis of the limitations of some traditional models (such as ARIMA, Kalman Filter) and simple neural network models was not in-depth enough. The specific deficiencies of these models in processing traffic flow data were not elaborated in detail, nor were there specific aspects in which the BiLSTM model could significantly improve the prediction results, such as the comparison of prediction performance in different traffic flow patterns and different time periods. It is suggested to further conduct an in-depth analysis of the advantages and disadvantages of the traditional model and the BiLSTM model in traffic flow prediction. Demonstrate the performance differences of different models when dealing with different traffic conditions and data characteristics through specific cases or experimental results, highlighting the advantages and applicable scenarios of the BiLSTM model.
2. There is a lack of in-depth explanation and analysis of the prediction results of the BiLSTM model. It does not fully explain how the model learns the patterns and characteristics of traffic flow from multi-source data, as well as the specific influence mechanism of spatial detector interaction on the model prediction. These make it difficult for readers to understand the internal logic and source of advantages of the model. It is suggested to add the interpretation and analysis of the BiLSTM model, and explore the key factors and mechanisms of the model when learning traffic flow patterns, such as how the model captures the temporal dependence and spatial correlation of traffic flow, and how to improve the prediction accuracy through multi-source data fusion. A more detailed explanation can be provided in combination with the structure and parameters of the model.
3. In the experimental section, although tests of different input combinations and spatial interactions were conducted, the design of the experiment and the presentation of the results could be clearer and more systematic. For example, when demonstrating the impact of different input combinations on the prediction accuracy, a more intuitive chart form can be adopted to present the results of all detectors in the westbound and eastbound directions, so as to facilitate readers' comparison and analysis.
4. The manuscript mainly focuses on the improvement of the prediction accuracy of the model, but there is less discussion on the application value and potential limitations of these improvements in actual traffic management. For instance, how to combine the prediction results with actual traffic management measures such as traffic signal control and traffic information release, as well as issues such as real-time performance, data quality, and model update that may be faced in practical applications, has not been deeply discussed. It is suggested that discussions on the practical application value and limitations be added in the manuscript. Cooperation or consultation with traffic management departments or experts in related fields should be carried out to explore how to better apply the research results to actual traffic management, as well as the possible problems and solutions encountered in the application process. So as to improve the practicality and guiding significance of the manuscript.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf