Application of Modified Machine Learning Technique in Traffic Safety Analysis

Abstract

Stop-sign compliance remains a persistent safety challenge in high-density, low-speed urban environments such as university campuses. This study investigates the predictors of traffic control compliance at an enclosed urban college campus using a Modified Machine learning technique (three-stage Stepwise Random Forest and Logistic Regression (SRFLR) framework). Observational data from 465 drivers were analyzed to move beyond raw correlations and capture the complex interactions between spatial, social, and demographic factors. The final predictive model identifies a hierarchy of influence dominated by directional intent and situational risk assessment. Direction of travel emerged as the most significant predictor, as drivers leaving the campus exhibited a greater likelihood of compliance, indicating that the purpose of travel plays a critical role in shaping driving behavior. Social and physical hazards also served as primary "forcing functions," as the presence of opposing traffic and pedestrians were consistent predictors of a full stop. While driver gender was a significant factor in initial screenings, its influence was reduced when controlling for environmental risks, indicating that demographic variances are partially explained by situational travel patterns. These findings suggest that campus safety interventions should prioritize engineering environments that naturally reinforce risk perception through enhanced visual salience and social monitoring. Such findings should also apply to situations beyond college campuses.