Intelligent Traffic Prediction in Smart Transportation Systems: A Comparative Review of Machine Learning and Deep Learning Approaches

Abstract

In recent years, ML-based models have garnered significant attention from both the automotive industry and academia for their potential to enhance Internet of Vehicles (IoV) systems. By accurately predicting traffic and road conditions, these models enable various safety and infotainment applications to optimize network resources and improve the overall quality of service. Real-time traffic flow forecasting also plays a crucial role in enhancing the efficiency of topology control and mobility management protocols within IoV networks. [1] However, despite the ongoing focus on improving prediction accuracy, an essential question remains unanswered: Are ML-based prediction models suitable for real-time traffic prediction? Addressing this requires a comprehensive study to evaluate the efficiency of these models. This article examines the effectiveness of several ML-based traffic flow prediction schemes by analyzing both their predictive accuracy and computational time requirements. Through a detailed quantitative analysis, we identify key factors that may limit the practical deployment of these models for real-time applications in IoV environments.

Introduction

This paper reviews machine learning (ML) models for real-time traffic prediction in Intelligent Transportation Systems (ITS) and the Internet of Vehicles (IoV). Accurate traffic forecasting is essential for improving road safety, reducing congestion, optimizing route planning, and enabling adaptive traffic signal control. However, deploying ML models in real time requires balancing prediction accuracy with computational efficiency and response speed.

Framework for Traffic Prediction

The proposed framework consists of four stages:

1. Data Acquisition – Collecting traffic data from sources such as loop detectors, GPS devices, and cameras.
2. Data Preprocessing – Cleaning data, normalizing values, and handling missing information.
3. Model Inference – Applying classical and deep learning models to predict traffic conditions.
4. Deployment – Integrating predictions into ITS applications such as congestion management, route guidance, and adaptive signal control.

A continuous feedback loop updates predictions based on real-time traffic conditions.

Research Contributions

The study:

• Compares both traditional and deep learning traffic prediction models.
• Evaluates not only prediction accuracy but also computational feasibility, scalability, and robustness.
• Uses benchmark datasets such as METR-LA and PEMS-BAY for performance comparison.
• Highlights the importance of balancing accuracy with real-time deployment requirements.

Evaluation Metrics

Traffic prediction models are assessed using:

• Mean Absolute Error (MAE) – Average prediction error.
• Root Mean Square Error (RMSE) – Penalizes larger prediction errors.
• Mean Absolute Percentage Error (MAPE) – Measures percentage error.
• R-Squared (R²) – Indicates how well the model explains traffic variability.
• Inference Time – Time required to generate a prediction.
• Latency and Throughput – Measures responsiveness and processing capacity.

Models Reviewed

Traditional Models

• Linear Regression – Fast and simple but limited in handling complex traffic patterns.
• Decision Trees – Efficient but prone to overfitting.
• Support Vector Machines (SVMs) – Accurate but computationally expensive for large datasets.

Advanced Models

• Long Short-Term Memory (LSTM) Networks – Effective for sequential traffic prediction but require significant computational resources.
• Graph Neural Networks (GNNs) – Capture spatial relationships in road networks but are difficult to scale for large cities.
• Reinforcement Learning (RL) – Useful for adaptive traffic signal control but requires extensive training and simulations.

Key Findings

• Deep learning models such as LSTMs, GNNs, and advanced graph-based approaches generally provide higher prediction accuracy than traditional methods.
• Simpler models like Linear Regression and Decision Trees offer faster inference times (1–5 ms) and lower memory requirements, making them suitable for edge devices.
• Advanced models may require 50–200 ms or more per prediction, limiting their real-time applicability without powerful hardware.
• Robustness to noisy or incomplete data is generally higher in deep learning models, while simpler models often require additional regularization.

Conclusion

In this research, we examined several machine learning models used for traffic prediction, such as deep learning architectures, reinforcement learning techniques, and conventional statistical methods. In this study, we considered the performance and limitations of these models in terms of accuracy, computational performance, scalability, and real-time application. This study highlights the significance of striking a balance between deployment practicalities and predictive performance by incorporating benchmark results, runtime comparisons, and standardized evaluation metrics like MAE, RMSE, and inference time. Despite the high accuracy of models like LSTM and GNN, their computing requirements may prevent real-time use, exposing a disconnect between academic achievement and practical application. Future work should concentrate on creating edge-deployable and hybrid solutions that maximize accuracy and runtime, as well as verifying these models in dynamic, realworld traffic situations. Machine learning-based traffic prediction is still a vital tool for creating intelligent and effective transportation systems, even as urban mobility issues increase. This review contributes a performance centered, deployment-aware lens to traffic prediction research—bridging theoretical advances with practical applicability.

References

[1] P. Sun, N. Aljeri, and A. Boukerche, “Machine learning-based models for real-time traffic flow prediction in vehicular networks,” IEEE Network, vol. 34, no. 3, pp. 178–185, 2020. [2] Y. Li, R. Yu, C. Shahabi, and Y. Liu, “Diffusion convolutional recurrent neural network: Data-driven traffic forecasting,” in International Conference on Learning Representations (ICLR), 2018. [3] J. Wu, Y. Tan, and B. Ran, “Short-term traffic flow prediction using a hybrid deep learning model,” Transportation Research Part C: Emerging Technologies, vol. 107, pp. 115–129, 2019. [4] W. Huang, G. Song, H. Hong, and K. Xie, “Deep architecture for traffic flow prediction: Deep belief networks with multitask learning,” IEEE Transactions on Intelligent Transportation Systems, vol. 15, no. 5, pp. 2191–2201, 2014. [5] X. Ma, Z. Tao, Y. Wang, H. Yu, and Y. Wang, “Long short-term memory neural network for traffic speed prediction using remote microwave sensor data,” Transportation Research Part C, vol. 54, pp. 187–197, 2015. [6] H. Yu, Z. Zheng, and X. Gao, “A survey on the applications of reinforcement learning in vehicular networks,” IEEE Access, vol. 8, pp. 12430–12449, 2020. [7] Z. Cui, R. Ke, and Y. Wang, “Deep bidirectional and unidirectional lstm for network-wide traffic speed prediction,” arXiv preprint arXiv:1801.02143, 2018. [8] J. Jiang et al., “PDFormer: Propagation Delay-Aware Dynamic Long-Range Transformer for Traffic Flow Prediction,” arXiv preprint arXiv:23s1.07945, 2023. [Online]. Available: https://arxiv.org/abs/2301.07945 [9] M. Yaqub et al., “Predicting Traffic Flow with Federated Learning and Graph Neural with Asynchronous Computations Network,” arXiv preprint arXiv:2401.02723, 2024. [Online]. Available: https://arxiv.org/abs/2401.02723 [10] C. Chen et al., “Gated graph sequence neural networks for traffic forecasting,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 33, 2019, pp. 485–492. [11] M. Leinonen et al., “Long Short Term Memory Based Traffic Prediction Using Multi-Source Data,” International Journal of Intelligent Transportation Systems Research, vol. 23, pp. 354–371, 2025. [12] C. Meese et al., “BFRT: Blockchained Federated Learning for Real-time Traffic Flow Prediction,” arXiv preprint arXiv:2305.17677, 2023. [Online]. Available: https://arxiv.org/abs/2305.17677

Copyright

Copyright © 2026 Smriti Nayak, Sonali Ganiga, Sharvari Poojary, Harish Kunder. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.