Fatigue driving is one of the important factors causing many traffic accidents. As a signal that directly reflects the electrical activity of the brain, brain electricity has become the focus of research to assess the detection and early warning of driving fatigue. This paper proposes a combinatorial experimental method, which carries out independent component analysis and analysis for different subjects, and then performs feature extraction of sample entropy, information entropy, fuzzy entropy and AR coefficient. Finally, an iterative algorithm is applied to minimize the least squares vector machine. Three different core classifiers are integrated into one strong classifier. At the same time, this paper discusses the AR order in feature extraction stage and compares the effects of three different nuclear classifiers. The average recognition rate of experimental results is 93%, which proves the advantages of this method and passes the 50% cross-validation accuracy rate of 91.04%, which verifies the robustness of the method and promotes safe driving assistance based on EEG signals to some extent. Monitoring system research.