I commend the authors for their insightful work, “Multicriteria Assessment Method for Network Structure Congestion Based on Traffic Data Using Advanced Computer Vision.” This study addresses a significant and timely issue, offering valuable contributions to understanding and managing traffic congestion using modern computational techniques. The integration of computer vision with machine learning algorithms, such as LSTM, GRU, and vision transformers, is commendable, and the proposed framework is innovative. Additionally, the focus on multicriteria decision analysis (MCDA) to optimize traffic flow enhances the practical applicability of the work.
However, certain aspects of the study warrant further scrutiny to enhance its robustness and transparency. The methodology, while innovative, raises some concerns, particularly regarding data selection and parameter justification. For example, the reliance on GPS data from specific partner organizations without detailed information on data diversity or representativeness could introduce biases. Moreover, while the A* and Viterbi algorithms are appropriately selected for route optimization and state estimation, the paper does not sufficiently discuss their limitations, such as computational complexity in large-scale networks or potential inaccuracies in noisy datasets. A deeper discussion of these constraints and comparative analysis with alternative algorithms would strengthen the study.
The use of neural networks for traffic prediction and weather condition forecasting is ambitious but lacks empirical validation through comparative benchmarks. For instance, the reported prediction accuracy of 86% for precipitation and mAP > 90% for vehicle detection could benefit from validation against alternative models or datasets to demonstrate superiority or generalizability. Similarly, the hybrid use of optical flow with vision transformers, while intriguing, lacks detailed technical exposition, making replication and evaluation challenging.
The discussion of criteria for decision-making and their assigned weights is detailed but might oversimplify nuanced factors influencing traffic congestion. For instance, the weights assigned to road surface quality and weather conditions seem low compared to their practical impact in real-world scenarios. Additionally, while the authors commendably incorporate expert opinions for qualitative criteria, the potential subjectivity and variance in expert judgment are not sufficiently acknowledged or mitigated.
The cited literature is extensive and relevant; however, the manuscript would benefit from deeper engagement with foundational works on traffic modeling and decision-making frameworks. Furthermore, while the visualizations and figures are clear, some, such as the depiction of network structures, lack the depth required to fully understand the proposed methodology’s mechanics. A more structured presentation of the workflow, alongside detailed sensitivity analyses, could enhance clarity and accessibility.
Addressing these methodological and contextual gaps, alongside greater transparency in data and algorithmic implementation, would significantly bolster the manuscript’s rigor and impact. These refinements would enable the study to make a stronger and more enduring contribution to traffic management research using computational intelligence.