Driver fatigue is a major contributor to road accidents, leading to reduced attention, slower reaction times, and impaired decision-making. This study presents a multi-level fatigue detection framework based on electroencephalography (EEG) signals, in which a Convolutional Neural Network (CNN) is employed to extract spatial patterns, and a Long Short-Term Memory (LSTM) network is used to model temporal dynamics in a cascaded architecture. To handle the high dimensionality and redundancy of EEG data, Compressed Sensing (CS) is applied with various compression ratios. Experimental results demonstrate that the proposed system achieves over 90% accuracy and an F1-score above 90% in multi-level fatigue classification. Even at a compression ratio of CR = 40%, the accuracy remains above 90%, while reducing the data volume by approximately 40%. Additional analyses using sensitivity, specificity, Cohen’s kappa, and ROC curves confirm the superiority of the proposed approach compared to baseline models (without CS or with simpler architectures). These findings indicate that the proposed framework is well-suited for real-time, portable driver monitoring systems.
