This paper provides a method for the classification of five unique sleep stages using the newly proposed DREAMT dataset (Dataset for Real-Time Sleep Stage Estimation using Multisensor Wearable Technology). With the use of photoplethysmography (PPG) signals alone, the approach here obtains record-breaking classification accuracies for balancing the data through downscaling (balanced dataset 1) as 99.35% for healthy subjects, 100% for snoring patients, 100% for subjects with excessive daytime sleepiness (EDS), and 98.82% for obstructive sleep apnea (OSA) patients and as 97.38% for healthy subjects, 97.73% for snoring patients, 98.84% for subjects with excessive daytime sleepiness (EDS), and 94.68% for obstructive sleep apnea (OSA) patients when the data is balanced using Synthetic Minority Oversampling Technique (SMOTE). This study is a notable achievement as the first to achieve such a high degree of precision in five-stage sleep categorization. Its success relies on a groundbreaking architecture with a combination of convolutional neural networks (CNN) and gated recurrent units (GRU), providing stable performance across a wide variety of patient categories. Future research will focus on combining the proposed model with wearable technology, generalize it to multi-sensor fusion, and enhance its generalizability to the broad classification of sleep disorders.