Enhance statistical features with changepoint detection for driver behaviour analysis

Driver behaviour modelling is a critical field that addresses complex and dynamic driving behaviours on roads with the goal of enhancing road safety, reducing air pollution, and improving vehicle performance. Recent advancements in sensor technology and machine learning (ML) techniques have facilitated the capture and analysis of driver behaviour patterns. Nonetheless, the efficacy of ML models heavily relies on the quality of the data used. Therefore, developing feature extraction techniques that provide high-quality inputs is crucial. In this paper, we conceptualised, implemented, and evaluated a novel feature model called Changepoint-based Statistical Feature (C-bSF). Initially, we extracted various statistical functions from raw sensor data, which were then aggregated using lagging windows. Following this, a changepoint detection method was used to derive the C-bSF feature. We compared the performance metrics of this new approach with other feature extraction methods, demonstrating the superiority of C-bSF in driver behaviour classification tasks across three datasets.