Development of a risk assessment tool based on driver behavior and environment

Abstract

Driver error is the leading cause of vehicle crashes; roadway segment characteristics and environmental conditions follow. However, no safety prediction model has been introduced to estimate risk associated with driver behavior, roadway segment, and environmental conditions. In this research, a probabilistic risk-based model for each individual driver with respect to driver demographic (age and gender), behavior, roadway characteristics, and weather condition is introduced. In the proposed model, the driver is placed in either no-crash, near-crash or crash categories for a given time stamp. Multinomial Logistic Regression (MLR) approach is used for estimating the risk (odds ratios) using crash and near-crash data, as well as normal driving data. The 100-car naturalistic driving study data sets are used to develop the model. The developed method demonstrates reliable performance in detection of outcome category.