Evaluating Deep Learning Architectures for EEG-Based Driver Fatigue Detection: A Comparative Analysis

Driver fatigue is a leading cause of road accidents and fatalities worldwide, highlighting the urgent need for effective and reliable detection systems. Recent advancements in deep learning (DL) have demonstrated superior capabilities in electroencephalogram (EEG)-based driver fatigue detection. However, existing methods often struggle with the accurate classification of both spatial and temporal features of EEG signals, necessitating significant improvements to enhance detection reliability and performance. This study presents a comparative analysis of deep learning models for EEG-based driver fatigue detection. To process EEG data, the Continuous Wavelet Transform (CWT) is used to generate time-frequency images, effectively capturing both spectral and temporal information. These images are then classified using different DL models, including single classifiers—Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Vision Transformer (ViT), Visual Geometry Group 16 (VGG-16), and Gated Recurrent Unit (GRU)—as well as hybrid classifiers such as CNN-LSTM, CNN-GRU, and CNN-ViT. The proposed approach is validated using a publicly available EEG dataset specifically collected for driver fatigue detection. Experimental results demonstrate that the single classifier CNN and the hybrid classifier CNN-LSTM achieve superior accuracies of 97.9% and 98.34%, respectively. These findings highlight the potential of these models in building reliable driver fatigue detection systems, contributing to enhanced traffic safety and reduced accident rates.