Automated Performance Evaluation of Multiple Classification Models with Feature Selection for EEG-Based Driver Fatigue Detection

This paper presents a detailed comparative analysis of 34 classification models for detecting driver fatigue based on EEG signals. The performance of the models is evaluated based on accuracy, training time, and error rate by taking all and most informative features derived from the EEG dataset. The publicly available well-known 40-channel EEG data with ground truth of 12 healthy candidates are used for experimental work. Candidate-independent classification is accomplished by utilizing 11 significant features after extracting them from EEG signals. This research also investigates the performance of three feature selection algorithms: Minimum Redundancy Maximum Relevance (MRMR), chi-square, and ReliefF for classifying each candidate's normal and fatigue states. Results indicate that the fine Gaussian SVM classification model is superior in accuracy (87.9%), whereas Linear Discriminant analysis has the least training time (1.76 Sec) with <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{PSD} \beta \mathbf{1},\ \mathbf{Std} \alpha$</tex>, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{Std} \beta$</tex> are the most informative features.