Predictability of driver’s stop/go decisions at flashing-light-controlled grade crossings by coupling functional brain network and deep learning methods

Detecting and predicting the stop/go decisions of drivers at grade crossings is crucial for enhancing road safety. Electroencephalography (EEG) data, which provides direct and effective physiological indicators for recognizing driver states, combined with associated machine-learning techniques, can be used to monitor driver decisions. However, the ability of EEG to predict a driver’s stop/go decisions remains unclear. To investigate this, we collected both EEG and behavioral data from drivers at a flashing-light-controlled grade crossing, where stop/go decisions are critical, using a driving simulator. Herein, we propose an EEG-based prediction framework that combines functional brain network analysis with conventional neural networks (FBN-CNNs) to predict drivers’ stop/go decisions. The functional brain network was measured using phase-lag index matrices and minimum-spanning tree techniques. We subsequently compared the obtained results of the FBN-CNN with those from traditional machine learning methods, specifically random forest (RF) and Support Vector Machines (SVM). The results indicate that when facing a flashing red light, drivers who decide to stop exhibit stronger alpha band connectivity and weaker delta and theta activity than those who run the red-light. Furthermore, the FBN-CNN model outperformed the machine learning methods (RF and SVM) in both extracting EEG features and achieving high prediction accuracy. Interestingly, the EEGs of drivers during normal driving stages could help to predict their stop-or-go behavior at the onset of a flashing red light. In the typical dilemma zone, combining EEG data from the normal driving stage with those from the pre-decision stage improved the accuracy from 76% to 90%. These findings demonstrate the efficacy of EEG and deep learning methods in driver decision monitoring.