L3-AVs Conflict Risk Prediction in Highway Maintenance Areas Using the CatBoost Model and SHAP Method Explanation

The study focuses on takeover and conflict risk of Level 3 autonomous vehicles (L3-AVs) in highway maintenance areas. Analysis of autonomous vehicle collisions shows that many of them are related to takeover process and collisions occur more frequently on highways. However, existing studies lack analysis of L3-AV performance in highway maintenance areas. In this study, we investigated the traffic flow and maintenance situation of S68 highway in Zhenjiang City and simulated it in SUMO, then compared the L3-AV takeover details based on length and number of lanes of maintenance area, fitted the prediction of the conflict data using the CatBoost model, and interpreted the prediction results in terms of global and local features by the SHAP theory. The results show that the length of the maintenance area and the number of lanes influence both the initial takeover and the takeover frequency. The relative speed between the L3-AV and the surrounding vehicles plays an important role in the conflict likelihood during traveling in the maintenance area. The findings of this paper are important for optimizing highway maintenance area configurations and developing L3-AV conflict avoidance techniques in specific scenarios.