EEG-based Prediction of Driver Takeover Performance

Abstract

In the context of conditional autonomous driving, ensuring a safe takeover is of paramount importance. While previous studies have delved into factors influencing drivers’ takeover performance, there remains a gap in research concerning the development of performance models capable of predicting takeover quality. To address this challenge, this study focuses on predicting driver takeover performance before the issuance of a takeover request based on Electroencephalogram (EEG) features. For this purpose, 72 subjects were recruited to participate in a driving simulation experiment, responding to a total of eight takeover events. Both their EEG signals and driving performance data were recorded. The takeover performance was subsequently categorized as high, medium, or low quality through a subjective review of the takeover process videos. A total of 480 EEG features, such as the power of α band, were extracted. Five machine learning models: Decision Trees (DT), Support Vector Machine (SVM), Random Forest (RF), Light Gradient Boosting Machine (LightGBM), and Multi-layer Perceptron (MLP), were utilized to develop the takeover performance prediction models. The results showed that the LightGBM model outperformed others, achieving an accuracy of 84.2% and an F1 score of 83.0%. In contrast, the DT model demonstrated the lowest performance, with an accuracy of 59.4% and an F1 score of 57.8%. This study underscores the potential of machine learning models in predicting driver takeover performance, thereby contributing to the advancement of machine learning applications in the field of autonomous driving.

Keywords: Autonomous driving, Takeover performance, EEG, Machine learning, Predictive modeling

DOI: 10.54941/ahfe1005233