Advances in Driver Behavior Classification: Traditional Paradigms, Deep Learning, Generative AI, and Explainable Systems

The classification of driver behavior has emerged as a central research priority in the pursuit of safer and more intelligent transportation systems. This review examines the evolving role of machine learning (ML) in modeling driver behavior, tracing its historical foundations, surveying current methodologies, and outlining prospective advancements. The application of machine learning models, including Support Vector Machines, Artificial Neural Networks, Decision Trees, and Deep Learning architectures, has proven instrumental in recognizing both normal and risky driving behaviors across a range of contextual conditions. Recent developments in ensemble learning, neuro-fuzzy systems, hybrid neuro-genetic approaches, and generative AI techniques such as Generative Adversarial Networks (GANs), Variational Auto-Encoders (VAEs), and diffusion models have significantly broadened the scope and fidelity of behavior recognition systems. Attention-based transformer models and large language models (LLMs) introduce semantic reasoning to multimodal driver data, improving interpretability and robustness. The paper emphasizes explainable AI (XAI), employing both model-specific and model-agnostic methods to enhance transparency and trust in classification outputs. It also addresses key technical and methodological challenges data imbalance, sensor noise, model interpretability, and real-time processing constraints. It proposes potential solutions grounded in continual learning, multimodal fusion, and federated learning. The integrating XAI with learning paradigms and emerging digital infrastructure to propose a context-aware, personalized, and ethically aligned roadmap for adaptive, transparent driver behavior classification in next-generation intelligent transportation systems.