Understanding and Predicting Drivers' Seatbelt Usage in Crashes in Virginia

Abstract

Despite the efforts of governmental and nonprofit agencies to increase seatbelt usage in the state of Virginia, drivers continue to drive while unrestrained. A better understanding of drivers' seatbelt usage would allow government officials and nonprofit agencies to more effectively target the right locations and populations with enforcement activities and education programs aimed at reducing unrestrained crashes. Recent literature has focused on identifying factors (such as sociodemographic characteristics of drivers) that correlate with seatbelt usage. This work aims to discover additional characteristics of unrestrained crashes and to predict the occurrence of unrestrained crashes in Virginia. To achieve these objectives, inferential analysis and predictive modeling were performed on Virginia crash data collected during the 2015 through 2017 time period and the seatbelt conviction data for these drivers. For the inferential part, hypothesis testing methods were used to uncover significant relationships between variables. Kernel density estimation (KDE) was used to identify spatial and temporal differences in restrained versus unrestrained crashes. For the predictive part, predictive machine learning models such as logistic regression and random forests were built to predict whether a crash was restrained or unrestrained. Results from this study will aid governmental and other agencies to develop occupant protection programs, increase public awareness, and target education and enforcement activities.