Factors affecting traffic risks on bridge sections of freeways based on partial dependence plots

Most of the safety studies mainly takes the probability of crashes or relevant records as the main indicators. Accident-based safety analysis may lead to overestimation or underestimation of traffic risks, which are delayed and sparse. Aggressive driving behavior is one of the major causes of traffic accidents, which can accurately measure traffic risk, but is not widely adopted due to limitations in data collection. This paper took the Traffic Order Index (TOI) as the surrogate index of safety risk based on aggressive driving behavior and speed variation and developed the Multinomial Logistic Regression (MLR) and the Random Forest (RF) model to identify risk level on bridge sections of freeway, which can minimize the restrictions of crash occurrence or crash-related data in discovering contributing factors of traffic risks. The results revealed that the RF has a better performance than MLR in the performance comparison of the two classifiers. The feature importance based on the Gini coefficient was used to identify the most influential variable of identified results of risk. The top four ranked variables that significantly affect the identified results of traffic order level are congestion index, road section types, the level of the number of users, and weather. In addition, the partial dependency plots of the explanatory variables are presented to reveal interactions between different variable types on traffic risks. Finally, the conclusion based on the traffic order level analysis has basically corresponded to the accident analysis. Identification of these specific risk prone conditions could improve our understanding of traffic risk and would shed light on countermeasures for improving the safety of bridge sections of freeways.