Decision-adjusted driver risk predictive models using kinematics information

Accurate prediction of driving risk is challenging due to the rarity crashes and individual driver heterogeneity. One promising direction tackling this challenge take advantage telematics data, increasingly available from connected vehicle technology, obtain dense predictors. In work, we propose a decision-adjusted framework develop optimal models using telematics-based behavior information. We apply proposed identify threshold values for elevated longitudinal acceleration (ACC), deceleration (DEC), lateral (LAT), other model parameters predicting risk. The Second Strategic Highway Research Program (SHRP 2) naturalistic data were used with decision rule identifying top 1% 20% riskiest drivers. results show that improves precision by 6.3% 26.1% compared baseline non-telematics superior based on receiver operating characteristic curve criterion, 5.3% 31.8% improvement in precision. confirm thresholds ACC, DEC LAT are sensitive rules, especially when small percentage high-risk This study demonstrates value kinematic crash necessity systematic approach extracting features. method can benefit broad applications, including fleet safety management, use-based insurance, intervention, as well connected-vehicle technology development.