A hybrid exploratory approach for understanding risk driving behaviors of bus drivers: A case study of Nanjing, China

Abstract

Risk driving behaviors among bus drivers raise growing concerns for public transportation operations, and identifying key influential factors can improve this situation. Based on 117,859 actual operation records from No. 851 bus line in Nanjing, causal relationships between five types of risk driving behaviors and influence factors were investigated by a framework of binary logit models to capture unobserved group and individual heterogeneities. Then, a random forest based SHAP model was utilized to provide further insights into potential inconsistencies. The empirical findings demonstrate that the performance of fixed effect binary logit models is consistent with that of random forest, as well as between the random effect and random parameter binary logit models. Besides, high correlations between land departure, vehicle proximity, and forward collision are observed. Further, travelling speed is identified as the predominant risk indicator, with lower speed being the determinant for distraction driving. Interestingly, the probability of forward collision increases beyond the distance of 50 m from bus bay entrances, and fatigue driving is more prone to occur at the locations less than 50 m from bus bay exits. Specifically, fatigue driving is mainly attributed to temporal and road environment characteristics, and distraction driving is more likely to happen on the single-lane roads with sharp acceleration and deceleration. Moreover, correlations between unobserved heterogeneities and some intervention measures for specific risk driving behaviors are quantified and proposed. Current findings could provide empirical evidence for implementing road safety measures and strategies in public transportation, and serve as supporting evidence for designing safety training programs for bus drivers.