Driving behavior profiling analysis and management based on Internet of Vehicles (IoV) Big Data platform

Abstract

The Internet of Vehicles (IoV) turns every connected automobile into a rolling sensor, delivering terabytes of time-stamped kinematic data that can illuminate how people actually drive. This study puts forward a scalable analytic framework that translates 3.7 million raw driving events recorded by 24 000 vehicles over fifteen months (2022-2023) into an interpretable taxonomy of driving personas. By fusing density-based noise removal with hierarchical agglomerative clustering, six recurrent driver profiles emerged, each marked by distinctive patterns in acceleration, braking, cornering and compliance with speed limits. The subsequent random-forest classifier achieved 87.3% accuracy $(F1=0.842)$ in real-time profile assignment. Field deployment of the profiling engine, coupled with personalised feedback, cut high-risk manoeuvres by 27.8% and squeezed 32.4% more kilometres out of every litre of fuel. Temporal reliability $(ICC=0.824)$ and geographical portability (Pearson $r=0.791$ confirm that the profiles capture enduring behavioural styles rather than ephemeral moods. The framework therefore offers insurers, fleet managers and safety engineers a statistically sound basis for differential risk pricing, targeted coaching and context-aware driver-assist systems.