This study presents a hybrid model that combines Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) networks for automated sleep stage classification using EEG signals. The framework involves preprocessing, feature extraction, model training, and classification. EEG signals were cleaned using a 0.5–40 Hz band-pass filter to suppress noise while retaining key sleep-related information. Frequency-domain characteristics were derived through Fast Fourier Transform (FFT) and Power Spectral Density (PSD) to highlight relevant sleep patterns. The CNN extracted spatial features, whereas the BiLSTM captured temporal dependencies, facilitating the accurate classification of six stages: Wake (W), NREM (S1, S2, S3), and REM. The model, trained on the Sleep-EDF dataset, achieved 82.32% training accuracy and 79.88% validation accuracy. Evaluation metrics such as precision, recall, F1-score, and AUC-ROC confirmed strong overall performance, although some misclassifications occurred in the Wake and S2 stages. Hypnogram analysis further validated reliable detection of sleep transitions, with minor errors in rapid stage shifts. Future directions include optimization strategies, data augmentation, and attention-based mechanisms to enhance accuracy and strengthen applicability for sleep disorder diagnosis and monitoring.