Sleep disorders pose a serious health hazard, and hence, accurate classification is required for effective intervention at the right time. This study proposes an optimized architecture for deep learning from Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) to enhance the classification of sleep disorders. Now that the Sleep Health and Lifestyle Dataset has been made available publicly, the model uses advanced feature engineering, SMOTETomek class balancing, and hyperparameter tuning to reach the optimum performance level. The proposed architecture captures spatial and temporal patterns with convolutional and bidirectional LSTM layers, and dense layers with dropout regularization for preventing overfitting. Experimental results are significantly better compared to conventional methods with a classification accuracy of 95%, outperforming existing state-of-the-art methods. This paper demonstrates the potential of deep learning for automating sleep disorder diagnosis and improving clinical decision-making. Future studies will attempt to combine multi-modal data fusion with real-time application for improved diagnostic precision.