Are driving styles stable or context-dependent? New evidence from elderly and middle-aged drivers using simulator data

With the ageing population growing rapidly, the number of middle-aged and elderly drivers has surged correspondingly, sparking widespread concerns about their ability to maintain safe driving behaviour. To address safety concerns associated with the effect of age-related declines on driving performance, we perform driving simulator experiments to assess behavioural differences among drivers of different ages. This study involves 50 drivers aged 40–69 years who participate in 94 driving simulation tests on urban roads and motorways. The study investigates driving behaviour using a novel multi-modal driving style recognition algorithm. Utilising a comprehensive driving simulator dataset of middle-aged and elderly drivers, we systematically pre-process driving data, extracting key features such as speed, lane position deviation and vehicle heading movement. Temporal dynamics of driving behaviour are captured using a long short-term memory network, and probabilistic distribution modelling is performed using a Gaussian mixture model with the Wasserstein distance to quantify inter-individual variability. The obtained temporal and probabilistic features are subsequently fused and classified using an entropy-optimised decision tree, enabling robust differentiation between cautious and aggressive driving styles. Model evaluation incorporates information gain and cluster metrics analysis across diverse traffic conditions. We find that 36% of drivers display environment-related transitions in driving style (cautious ↔ aggressive), with such diverse driving styles being particularly prevalent among non-professional drivers, especially those who are middle-aged.