Safety critical event prediction through unified analysis of driver and vehicle volatilities: Application of deep learning methods

Transportation safety is highly correlated with driving behavior, especially human error playing a key role in large portion of crashes. Modern instrumentation and computational resources allow for the monitorization driver, vehicle, roadway/environment to extract leading indicators crashes from multi-dimensional data streams. To quantify variations that are beyond normal driver behavior vehicle kinematics, concept volatility applied. The study measures driver-vehicle volatilities using naturalistic data. By integrating fusing multiple real-time streams data, i.e., distraction, vehicular movements instability driving, this aims predict occurrence critical events generate appropriate feedback drivers surrounding vehicles. used which contains 7566 events, 1315 severe (i.e., crash near-crash), collected more than 3500 drivers. In order capture local dependency time-series 1D-Convolutional Neural Network (1D-CNN), Long Short-Term Memory (LSTM), 1DCNN-LSTM Vehicle volatility, impaired (in terms distraction) as input parameters. results reveal model provides best performance, 95.45% accuracy prediction 73.4% precision 95.67%. Additional features extracted CNN layers temporal between observations addressed, helps network learn patterns volatile behavior. can be monitor provide warnings alerts low-level automated vehicles, reducing their risk.