A comprehensive review of traffic prediction: From traditional machine learning to AutoML

In the rapidly urbanizing world, efficient traffic prediction is essential for reducing congestion, optimizing travel times, and enhancing road safety. Traditional machine learning (ML) models have long been used for traffic forecasting but often struggle with unstructured data and capturing the complex temporal and spatio-temporal relationships inherent in traffic networks. Deep learning (DL) models, by contrast, can effectively handle large datasets and learn complex patterns, yet they still demand substantial human expertise for architecture design, hyperparameter tuning, and dataset-specific adaptation. This paper presents a comprehensive review of the evolution of traffic prediction models, highlighting the limitations of ML and DL approaches and introducing Automated Machine Learning (AutoML) as a promising solution. We discuss how AutoML can automate key stages of the ML pipeline—including data preprocessing, feature engineering, model learning, and model updating—reducing the need for human expertise, improving generalizability, and enabling model adaptation across datasets. While some studies have integrated AutoML components into traffic prediction tasks, a fully automated, end-to-end pipeline remains an open research challenge. This review identifies current gaps, explores AutoML’s potential to address these challenges, and outlines future directions for advancing traffic prediction through AutoML.