Hybrid Machine and Deep Learning for Enhanced Sleep Disorder Diagnosis: A Scalable Approach for Real-Time Clinical Application

A. Rajesh1, Vaka Vineetha2, Shanmukha kumar karra3, Vemireddy Sai Subhash Reddy4,Lasya Lakshmi Putta5
1Assistant Professor, Dept of Information Technology, SV College of Engineering, Tirupati, India.
2B.Tech, Dept of Information Technology, SV College of Engineering, Tirupati, India.
3B.Tech, Dept of Information Technology, SV College of Engineering, Tirupati, India.
4B.Tech, Dept of Information Technology, SV College of Engineering, Tirupati, India.
5B.Tech, Dept of Information Technology, SV College of Engineering, Tirupati, India.

Abstract-Classifying sleep disorders such as obstructivesleep apnea and insomnia remains essential for enhancing health outcomes, yet existing machine learning approaches using the Sleep Health and Lifestyle Dataset face key limitations. Current systems apply preprocessingwith SMOTEENN for class imbalance, ANOVA hypothesis testing, Z-score scaling, and feature selection via Gradient Boosting Regressor-based Mean Decrease Impurity (MDI), evaluating 15 classifiers on original and engineered feature spaces (augmented with predictionsfrom seven base classifiers). While Gradient Boosting achieves 97.33% accuracy, 0.9733 precision/recall/F1 score, 0.9569 specificity, and 0.9953 AUC using five keyfeatures (Blood Pressure, BMI Category, Daily Steps,  Sleep Duration, Occupation), limitations include smalldataset size (374 samples) restricting generalizability, longer training times for high-accuracy ensemble models, reliance on original features over engineered ones, and lack of unsupervised methods or real-time deployment. This paper proposes an advanced hybrid frameworkaddressing these gaps by integrating unsupervised learning (e.g., clustering for anomaly detection), deep feature extraction via autoencoders, expanded multi source datasets, and optimized real-time lightweight models deployable on wearables. Benefits include improved robustness (targeting >98% accuracy), reduced training time by 50-70% through pruning and quantization, enhanced generalizability across demographics, and seamless integration into clinical systems for early diagnosis, minimizing manual polysomnography reliance and enabling scalable healthcare in resource-limited settings. Keywords: sleep disorders, Gradient Boosting, SMOTEENN, Mean Decrease Impurity (MDI), generalizability,unsupervised polysomnography