The purpose of this paper is to analyze the complex coupling relationships among accident factors contributing to the automobile and two-wheeler traffic accidents by establishing the Bayesian network (BN) model of the severity of traffic accidents, so as to minimize the negative impact of automobile to two-wheeler traffic accidents. According to the attribution of primary responsibility, traffic accidents were divided to two categories: the automobile and two-wheeler traffic as the primary responsible party. Two BN accident severity analysis models for different primary responsible parties were proposed by innovatively combining the Kendall correlation analysis method with the BN model. A database of 1560 accidents involving an automobile and two-wheeler in Guilin, Guangxi province, were applied to calibrate the model parameters and validate the effectiveness of the models. The result shows that the BN models could reflect the real relationships among the influential factors of the two types of traffic accidents. For traffic accidents of automobiles and two-wheelers as the primary responsible party, respectively, the biggest influential factors leading to fatality were weather and visibility, and the corresponding fluctuations in the probability of occurrence were 32.20% and 27.23%, respectively. Moreover, based on multi-factor cross-over analysis, the most influential factors leading to fatality were: {Off-Peak Period → Driver of Two-Wheeler: The elderly → Driving Behavior of Two-Wheeler: Parking} and {Drunk Driving Two-Wheeler → Having a License of Automobiles → Visibility: 50 m~100 m}, respectively. The results provide a theoretical basis for reducing the severity of automobile to two-wheeler traffic accidents.
Keywords: Bayesian network; Kendall rank correlation; automobile to two-wheeler traffic accidents; big data and traffic safety; severity of accidents.