Sleep disorders significantly impact human health and well-being, highlighting the need for accessible and efficient diagnostic tools. Traditionally, diagnosing these disorders involved uncomfortable overnight stays in sleep labs, limiting long-term monitoring and requiring expert oversight. Researchers have turned to wearable devices for automated sleep classification systems to overcome these challenges. Among these technologies, photoplethysmography (PPG) signals have emerged as a key component in developing automatic frameworks for multi-stage sleep classification. In this study, both machine learning (XGBoost) and deep learning techniques (LSTM, CNN) were employed to create models for classifying two-stage (Wake-Sleep) and three-stage (Wake-NREM-REM) sleep patterns. The proposed approach achieved an average classification accuracy of 87% and 76% for two and three stages, surpassing existing state-of-the-art models in the field. By utilizing wearable devices equipped with PPG sensors, individuals can now monitor their sleep patterns in real time, facilitating early detection and intervention for sleep disorders and enhancing overall health outcomes.