Roukos I., Garefalakis T., Michelaraki E., Yannis G., “Predicting driver behaviour in a cross-country naturalistic driving study using machine learning techniques”, Proceedings of the Transport Research Arena TRA 2026 Conference, Budapest, 18-21 May 2026.

As traffic incidents continue to pose serious public health risks, identifying and predicting unsafe driving behaviour has become increasingly important. This paper is an outcome of the European Union’s Horizon 2020 i-DREAMS project and applies the Safety Tolerance Zone (STZ) concept by classifying trip-level driving behaviour into three safety levels using naturalistic driving data. The aim of this work was to develop machine learning models in order to classify driver behaviour into three safety levels. To achieve this objective, a naturalistic driving experiment was conducted and data from Belgium and the United Kingdom were collected and analyzed. Variable importance was assessed using the Random Forest algorithm, resulting in the selection of nine key features. Four classification models were trained and evaluated through confusion matrices and standard performance metrics. SHapley Additive exPlanations (SHAP) values were used to interpret the contribution of each input variable, leading to the identification of CatBoost and LightGBM as the most effective models. Further SHAP analysis revealed that average speed was the most influential predictor across all safety levels, while harsh driving events played a crucial role in identifying dangerous behaviour. The findings highlighted the potential of interpretable machine learning models for real-time safety monitoring and suggest incorporating speed management and harsh event detection into future in-vehicle safety systems.