<p>Drowsiness detection using electroencephalogram (EEG) signals is a vital area of research with significant implications for road safety, occupational monitoring, and healthcare. This study presents a single-channel EEG-based drowsiness detection system leveraging advanced deep learning techniques. A single EEG channel is systematically selected based on prior research to identify the most informative signal. The Sleep-EDF dataset is utilized, with preprocessing steps including filtering and noise reduction. Initially, four recurrent neural network (RNN)-based architectures - LSTM, Bi-LSTM, GRU, and Bi-GRU are utilized with manually selected hyperparameters. Subsequently, Bayesian optimization is employed for fine-tuning, significantly improving model performance. Building upon this, the Bahdanau attention mechanism is integrated into all RNN architectures to enhance interpret-ability and classification accuracy. To further boost system performance and robustness, an ensemble approach is adopted by combining multiple Bahdanau-attention-enhanced deep learning models. The results are compared with existing deep learning approaches to evaluate improvements in accuracy and real-time detection capability.</p>