Next Article in Journal
Deep Multi-Component Neural Network Architecture
Previous Article in Journal
Eye Care: Predicting Eye Diseases Using Deep Learning Based on Retinal Images
 
 
Article
Peer-Review Record

Predicting Urban Traffic Congestion with VANET Data

Computation 2025, 13(4), 92; https://doi.org/10.3390/computation13040092
by Wilson Chango 1,*, Pamela Buñay 2, Juan Erazo 3, Pedro Aguilar 4, Jaime Sayago 1,*, Angel Flores 4 and Geovanny Silva 4
Reviewer 1:
Reviewer 2:
Computation 2025, 13(4), 92; https://doi.org/10.3390/computation13040092
Submission received: 27 January 2025 / Revised: 6 March 2025 / Accepted: 14 March 2025 / Published: 7 April 2025
(This article belongs to the Section Computational Engineering)

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper presents a machine learning approach for predicting traffic congestion using Vehicular Ad Hoc Network (VANET) data in Esmeraldas, Ecuador. The authors implement a Random Forest-based classification model that integrates multiple data sources and achieves high accuracy in congestion prediction. The study includes a detailed simulation using urban road segments and validates the approach through comprehensive performance metrics. I have some concerns as follows:

 

The study relies on data from a single urban area in Ecuador, which may limit its generalizability to other cities with different traffic patterns.

The significant imbalance in the dataset (96.38% non-congested vs. 3.62% congested) raises concerns about the model's robustness and the perfect accuracy scores further raise concerns about overfitting.

The paper could be better validated with real-world implementation data and data from different locations or traffic conditions. The evaluation also lacks the computational resources needed for the system's real-time implementation.

Limited consideration of weather conditions, special events, or other external factors that might affect traffic patterns.

The choice of RF needs more justification, as RF may struggle to capture complex temporal dependencies in traffic patterns, and DL models such as LSTM and Transformer have shown superior performance in time-series prediction tasks.

Author Response

  1. Study limited to a single urban area of Ecuador

In future research, we plan to broaden the scope of the analysis to include multiple urban areas in Ecuador and other regions. This will allow for a more robust comparison between cities with different congestion patterns, road infrastructure, and weather conditions.

  1. Imbalance in the dataset (96.38% non-congested vs. 3.62% congested)

To address this issue, the SMOTE (Synthetic Minority Over-sampling Technique) oversampling technique was implemented, which generates synthetic samples of the minority class to balance the dataset. This technique proved effective in improving the performance of the models by ensuring that both classes are represented more equitably.

  1. Improved validation with real data and diverse locations

In future research, we plan to implement a broader approach that includes the collection of real-time data from multiple urban locations. This will allow us to evaluate the robustness and generalization of the model in different contexts, as well as identify possible regional variations in congestion patterns. Additionally, the integration of data from IoT sensors, traffic cameras, and mobile applications will be explored to enrich the database and improve the accuracy of predictions.

  1. Weather conditions and external factors

In this study, a set of climatic and environmental variables that influence traffic and congestion patterns has been incorporated. These variables include:

Temperature: Affects driver behavior and road conditions (e.g., in extreme heat or cold).

Precipitable Water: Indicates the amount of water vapor in the atmosphere, which can influence the formation of rain and, therefore, traffic conditions.

Wind Speed: Can affect visibility and road safety, especially in strong wind conditions.

Atmospheric Pressure: Is related to changes in weather, such as the arrival of cold or warm fronts, which can influence traffic behavior.

Wind Direction: Can have a localized impact on the dispersion of pollutants or the formation of fog.

Relative Humidity: Influences driver comfort and the likelihood of fog or rain formation.

These climatic factors have been integrated into the model to improve its predictive capacity and understand how environmental conditions interact with urban traffic. In future studies, the inclusion of other external factors, such as special events, roadwork, and holidays, will be explored to further enrich the analysis.

  1. Choice of Random Forest (RF) and justification of other models like LSTM or Transformer

Advanced models, such as LSTM (Long Short-Term Memory) and Transformer, were used, which are particularly useful for capturing temporal dependencies in time series. After an exhaustive comparison, it was found that the LSTM model outperformed the Transformer in terms of accuracy and ability to predict congestion patterns over time. This is due to the sequential nature of traffic data, where current conditions heavily depend on previous conditions.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

  • Scientific Methods and Models

The authors used Random Forest algorithm to classify road section as congested or not. According to the numbers provided by the authors the dataset is heavily imbalanced and training classification models usually don’t perform perfect as the authors suggest with their results for such an imbalanced dataset as they described.

First the authors are advised to use oversampling techniques such as SMOTE to balance the dataset and then examine if they are training the model correctly because perfect classification metrics are rare.

Achieving perfect classification metrics is exceptionally rare. It usually hints at overfitting or potential data leakage, an overly simple or non-representative dataset or a methodological issue such as improper train test split or even training the model on a near duplicate of the target that would allow the model to understand the thresholds and always predict based on that. In addition, the confusion matrix if the model and its application is examined and verified should include also include numbers for each class.

Finally, authors are advised to enrich their research with a summarizing table of their literature review showcasing how similar research was used and justifying their choices, for example explain how the 10km/h threshold was chosen as an indication of traffic congestion.

Comments on the Quality of English Language

  • Language and Paper Format

Authors are advised to format some figures and tables in order to have more structured and formal appearance for example Figure 1 although is informing it lacks in terms of formatting. Phrases or words shouldn’t be cut and continued in the next line (example Manageme-nt). A resize of the figure should provide enough space to include entire words in each line.

The use of the PRISMA methodology is correct and authors made a good effort in explaining and describing the process for the literature review.

Additionally, authors should revise the text for some minor mistakes. For example, line 273 the text mentions a table with question marks instead of a number. Furthermore Table 6 has no caption authors should change the add caption text to the proper caption that describes the table and its information. In line 339 the authors repeat the same word “Precision” twice. The same is observed for the term “Recall”. Maybe the authors tried to categorize the metrics and then explain each one. In that case they should use bold or bullets instead of repeating the same word.

All tables and their legends should be in English there are some legends that include phrases that are not written in English, for example the legend in Figure 5.

Author Response

  1. Desequilibrio del conjunto de datos

Se ha identificado que el conjunto de datos presenta un alto nivel de desequilibrio, lo que a menudo afecta negativamente el rendimiento de los modelos de clasificación. Como han señalado los autores en sus resultados, los modelos de aprendizaje automático tienden a mostrar un sesgo hacia la clase mayoritaria cuando trabajan con conjuntos de datos altamente desequilibrados.

Para mitigar este problema y mejorar la representatividad de las clases minoritarias se ha implementado la técnica SMOTE (Synthetic Minority Over-sampling Technique). Esta metodología permite generar nuevas instancias sintéticas de la clase minoritaria, consiguiendo un conjunto de datos más equilibrado y favoreciendo un mejor rendimiento del modelo de clasificación.

  1. Evaluación de métricas y sobreajuste

Es poco frecuente obtener métricas de clasificación perfectas en tareas de aprendizaje automático. Los resultados demasiado altos pueden indicar un sobreajuste del modelo o incluso la presencia de fuga de datos, lo que compromete la validez de las predicciones.

Para abordar este problema y asegurar la generalización del modelo, se han implementado arquitecturas avanzadas de redes neuronales, en concreto Transformers y LSTM (Long Short-Term Memory). Estas arquitecturas permiten capturar dependencias temporales en los datos de tráfico y mejorar la precisión de las predicciones sin caer en problemas de sobreajuste.

  1. Revisión de la literatura y cuadro resumen

Para enriquecer la investigación y aportar una visión más estructurada de los estudios previos, se ha incorporado una tabla resumen que sintetiza los principales estudios relacionados con la detección de congestión vehicular. Esta tabla incluye aspectos clave como la metodología empleada, los hallazgos más relevantes y su impacto en la literatura existente. De esta forma, facilita la comparación de aproximaciones previas y contextualiza mejor la aportación del presente estudio.

  1. Formato de Figuras y Tablas

Se ha atendido la recomendación de mejorar la presentación visual de las figuras y tablas para que tengan una apariencia más estructurada y formal. En particular, se ha corregido la Figura 1, ajustando su formato y alineación, y se han realizado modificaciones en otras figuras y tablas para cumplir con los estándares de presentación académica. Estos ajustes buscan mejorar la claridad y comprensión de los datos presentados.

  1. Correcciones en el texto y numeración de las tablas

Se han revisado y corregido varios errores menores en el texto, incluidos problemas de redacción y referencias incorrectas. Por ejemplo, en la línea 273 se ha corregido una referencia incorrecta en la que un signo de interrogación reemplazaba el número de la tabla. Asimismo, se ha corregido la Tabla 6, que originalmente no tenía título, y ahora incluye una designación clara y coherente con su contenido. Además, se han revisado palabras repetidas y se ha mejorado la coherencia textual en diferentes secciones del documento.

  1. Traducción de Leyendas al español

Se ha realizado una revisión completa de las leyendas de las figuras y tablas para asegurar que todas estén escritas en inglés. Se identificó que algunas leyendas contenían frases en otros idiomas, como en la Figura 5, por lo que se realizó la traducción correspondiente para mantener la uniformidad y el estándar del documento.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

I thank the author for addressing some of my previous concerns. However, the current revision only provides a clean version without highlighting which parts were revised, and this information is also not included in the response, making it difficult to locate the modifications.

Additionally, the current version needs proofreading as there are typos, e.g., in Table 10. Some figures could also be redesigned to enhance readability and information delivery, e.g, Fig. 1.

Author Response

Comentarios 1: Además, la versión actual necesita una corrección porque contiene errores tipográficos, por ejemplo, en la Tabla 10. Algunas figuras también podrían rediseñarse para mejorar la legibilidad y la presentación de la información, por ejemplo, la Figura 1.

Respuesta 1: Se realizó un cambio completo en la estructura del artículo, incorporando las observaciones detalladas en la tabla 10 y figura 1. Adicionalmente, se implementó la técnica SMOTE (Synthetic Minority Over-sampling Technique) antes y después del procesamiento de los datos, lo que permitió balancear la distribución de clases en el conjunto de datos. Asimismo, se desarrolló un modelo predictivo utilizando redes neuronales, específicamente LSTM (Long Short-Term Memory) y Transformers, realizando una comparación entre ambos modelos para evaluar su desempeño y efectividad.

Reviewer 2 Report

Comments and Suggestions for Authors

The authors are advised again to correct their figures and translate all not English text to their proper English translation. For example, Figure 5 has Spanish in the legend, these should be clearly translated to English since the whole document is written in English.

Minor formatting and Layout adjustment in the Figures and Tables, as previously mentioned in the past review, all words should be in one line and not separated for example the last circle has the word cut in half.

Comments on the Quality of English Language

There are spelling mistakes in some Tables. For example, in Table 10 the authors use “Loos” instead of “Loss”.

Author Response

Comments 1: The authors are advised again to correct their figures and translate all not English text to their proper English translation. For example, Figure 5 has Spanish in the legend, these should be clearly translated to English since the whole document is written in English.

Response 1: The correction of the figures and their translation into English has been carried out, including Figure 5, which shows the integration of environmental data and the LSTM-Transformer networks for vehicular congestion prediction. This figure has been reviewed to ensure that the graphic elements and labels are correctly aligned and clear, both in its original version and in its English translation. Furthermore, it has been verified that the legends and titles are correctly translated and consistent with the content of the document.

Comments 2: Minor formatting and Layout adjustment in the Figures and Tables, as previously mentioned in the past review, all words should be in one line and not separated for example the last circle has the word cut in half.

Response 2:The graph has been corrected and the correction has been made in all tables to standardize the format, ensuring that the presented data are consistent and clear throughout the documentation. This includes the alignment of headers, uniformity in the use of fonts and styles, and the revision of legends and notes to ensure they are accurate and easy to understand.

Comments 3: There are spelling mistakes in some Tables. For example, in Table 10 the authors use “Loos” instead of “Loss”.

Response 3: All tables, including table 10, have been corrected to eliminate misspellings and ensure the clarity and accuracy of the information presented.

Back to TopTop