Risk evaluation of lane changing with time to collision and lateral acceleration

Lane-changing manoeuvres are a major contributor to motorway crashes, yet existing risk-assessment frameworks often overlook manoeuvre direction, lane-change type and spatial context. This study develops a data-driven framework for evaluating lane-change risk using the Next Generation Simulation I-80 trajectory data set. Time to collision (TTC), lateral acceleration (Ay), and vehicle kinematics were extracted, and an extreme gradient boosting (XGBoost) classifier was trained using risk labels defined as TTC <1.5 s or |Ay| > 2.5 m/s2. Five-fold cross-validation yielded an accuracy of 0.9978 ± 0.003, with precision, recall and F1-score of 1.00. TTC was approximately 1.4 times more influential than Ay in classifying risk. Rightward lane changes showed significantly higher individual risk than leftward manoeuvres, despite occurring less frequently. In addition, higher-speed freeway zones exhibited greater individual risk than high-volume merging areas near the on-ramp. Mandatory lane changes were 1.3 times more likely to be classified as high-risk than discretionary manoeuvres, although the difference was not statistically significant. These findings support the use of TTC-based thresholds, directional risk differentiation and spatial context in advanced driver-assistance systems and autonomous-vehicle safety systems.