Research

This study developed a takeover prediction model utilizing driving data from the autonomous bus, 'PantaG Bus'. To derive optimal prediction performance, sensitivity analysis was conducted by varying time window sizes rather than fixing them, and the performances of Logistic Regression, 1D-CNN, BiLSTM, and XGBoost models were comparatively analyzed. The results revealed that the XGBoost model with a 3-second window achieved the highest performance in terms of the F1-score and was thus selected as the optimal model. The model effectively detected takeover timing by capturing instantaneous and rapid change patterns in sensor data; specifically, it successfully predicted risks even in intersection dilemma zones and cut-in scenarios. Furthermore, feature importance analysis identified the vehicle's lateral behavior as the critical factor inducing control takeover. This study contributes to securing safety margins for the safe operation of autonomous buses and enhancing driver acceptance of the system, thereby suppressing unnecessary driver intervention during autonomous driving and improving operational efficiency.