Estimating the Vigilance of High-Speed Rail Drivers Using a Stacking Ensemble Learning Method

High-speed rail (HSR) drivers are the key part of operation safety, and their vigilance is the main factor affecting accident occurrence. Hence, an effective and reliable method to estimate HSR drivers' vigilance is needed to ensure driving safety. Given that drivers' reaction time can objectively and effectively reflect their vigilance, this paper proposed a two-layer stacking ensemble learning model to predict HSR drivers' reaction time to sudden stimuli based on electroencephalogram (EEG) signals. Three individual regression models were stacked together in the first layer to predict drivers' reaction time separately based on the inputted power spectral density features of EEG signals. Random forest was then used as the regression model in the second layer to negotiate the outputs from the first layer for a more accurate prediction of drivers' response time. The proposed model was trained and tested with the EEG data collected from 40 HSR drivers in a simulated driving experiment. The results show that the mean absolute error (MAE), root mean square error (RMSE), and goodness of fit ( R
²
) of the estimated reaction time when using our proposed model were 70.14(±13.02) ms, 102.19(±22.18) ms, and 0.74(±0.09), respectively, better than the corresponding results when using any of the regression models individually or comparing to six other popular methods. The impacts of EEG features from different brain regions and the individual differences between HSR drivers on vigilance estimation were also analyzed to further examine the performance of our proposed model.