Assessing Entry–Circulating Conflict Severity at Roundabouts in Mixed Traffic Using Post-Encroachment Time and Speed Profiles

Traffic roundabouts are generally associated with improved safety because of reduced conflict points and lower operating speeds; however, their performance under mixed traffic conditions in low- and middle-income countries remains inadequately understood. Safety assessment in such environments is constrained by heterogeneous traffic and limited availability of reliable crash data. Although surrogate measures of safety based on traffic conflicts provide a viable alternative, conflict severity is often defined using arbitrary thresholds. This study presents a data-driven framework to classify and predict conflict severity at urban roundabouts operating under mixed traffic conditions. Entry–circulating vehicle interactions were extracted from videographic data, and conflict severity was classified using k-means clustering based on post-encroachment time, entry speed, and circulating speed. The derived severity labels were used to develop a random forest model incorporating behavioral, traffic, and geometric variables. The model achieved an accuracy of 87.8% and an area under the receiver operating characteristic curve of 0.91. Entry speed and circulating speed emerged as the most influential predictors, followed by gap acceptance. The framework avoids predefined severity thresholds and supports proactive safety assessment within a Safe System perspective.