Feature Importance Estimation for Risky Driving Behaviour

This document is an extended abstract submitted for presentation at the Conference in Emerging Technologies in Transportation Systems (TRC-30). It focuses on feature importance estimation using clustering and classification techniques to analyze risky driving behavior. The introduction outlines the significance of dangerous driving behaviors, such as speeding and distractions, which contribute to traffic accidents. The methodology section details a proposed framework that processes raw data from vehicle telematics, employing K-means clustering to identify optimal feature combinations. The results indicate that distance to intersections and road curbs are critical in identifying risky behaviors. The document discusses the advantages of the proposed model over traditional methods, emphasizing its ability to assess feature importance transparently and comprehensively. Additionally, it acknowledges the computational challenges associated with the model and suggests potential solutions to enhance efficiency. Overall, the abstract presents a novel approach to understanding and mitigating risky driving behaviors through data-driven analysis.