MODELLING OF REAR END CONFLICT RISK AT UNCONTROLLED INTERSECTION USING SURROGATE SAFETY MEASURES

Abstract

Road traffic accidents are a global concern, with intersections being particularly vulnerable to collisions. Identifying the factors contributing to conflicts is crucial for predicting conflict risk and implementing effective measures to enhance vehicle safety. However, prior research on conflict severity has been based on aggregated data, overlooking individual vehicle dynamics. This study addresses this gap by utilizing trajectory data from individual vehicles to create a conflict risk classification model and identify the determinants of traffic conflicts. The study involves several stages, including cluster analysis of traffic conflict indicators, implementation of five machine learning classification models and interpretation of feature importance. Traffic conflict indicators such as MTTC, DRAC, and PSD were used to identify and classify conflict risk. Three clustering algorithms - K-means, spectral, and agglomerative - were employed to classify traffic conflict indicators into four risk levels: low, medium, high, and critical conflicts. Five machine learning models were evaluated: Logistic Regression, Decision Tree, Random Forest, XGBoost, and Support Vector Machine. The Random Forest algorithm outperformed all other models, achieving an accuracy of 91%, precision of 91%, recall of 92%, and an AUC score of 0.98. To address interpretability challenges in machine learning models, the SHAP analysis was employed to identify the significant variables and measure their impact on conflict risk. The study identified the top five features most likely to influence conflict risk, which included maximum deceleration of leader (MDL), standard deviation of spacing between vehicles (SDSP), maximum acceleration of follower (MAF), standard deviation of speed follower (SDSF), and mean longitudinal spacing between the vehicles (MSPA). Further analysis of the beeswarm and dependency plots for each risk class indicates that certain features, such as SDSF, MAF and MDL exhibit an increase in risk levels with increasing values, while a decrease in MSPA and headway (HW) is associated with increased conflict risk. These findings provide valuable insights for developing effective countermeasures to mitigate conflict risk and improve traffic safety, particularly for connected and automated vehicles utilizing advanced driver assistance systems