Driver Behavior Analysis in Simulated Jaywalking and Accident Prediction Using Machine Learning Algorithms

Road safety can be improved if traffic accidents can be predicted and thus prevented. The use of driver-related variables to determine the possibility of an accident presents a new analysis paradigm. We used a driving simulator to create a jaywalking scenario and investigated how drivers responded to it. A total of 155 valid participants were identified across demographics (age group and gender) and participated in the experiment. We collected driver-related data on eight types of perception/reaction times, vehicle-control data, accident occurrence data, and maneuvers used for obstacle avoidance. From the statistical analysis, it was possible to derive six variables with significant differences based on whether a traffic accident occurred. Furthermore, we identified the data's significant difference according to demographics. Artificial intelligence (AI)-classification models were used to predict whether an accident would occur with up to 90.6% accuracy. The data associated with the dangerous scenario obtained in this study were identified to predict the occurrence of traffic accidents.