Application of M5P model to examine crash frequency on the expressways in India

This study identifies the significant risk factors associated with crash frequency on expressways in India. For this crash data was collected from over 391 km of three expressways comprising a total of 13345 crashes. The expressways were divided into 981 segments based on uniform traffic and geometric characteristics, and traffic, geometry, and lighting conditions-related data from these segments was also gathered. The final dataset comprised 6577 data points, and it was analysed using M5P and random effect negative binomial (RENB) models. The results obtained from the modeling approaches were compared based on the correlation coefficient (CC), root mean square error (RMSE), and mean absolute error (MAE). Although both models performed well in identifying the risk factors, the crash prediction of M5P model was better than the RENB model. It provided simple linear equations for sets of conditions defined by the tree. The findings of M5P model also suggest that it better explains the effect of various independent variables on crash probability under different circumstances. Furthermore, sensitivity analysis utilising the M5P model indicates that it can generalise and predict crashes' physical events well. The findings of this study suggest that modification in the design of horizontal and vertical profiles, median openings, interchanges, minor accesses, and bus bays/truck laybys are needed to increase traffic safety on expressways in India.