Sleep Disorder Classification Using HSE-NET Ensemble: A Robust Machine Learning Framework

Abstract

Sleep regulation disorders, particularly insomnia and obstructive sleep apnea, can seriously affect mental well-being, emotional balance, and overall physical health. Although polysomnography is considered the standard clinical method for diagnosing these disorders, its high cost, time requirements, and dependence on specialized facilities make it difficult to access, especially in resource-limited healthcare settings. To address these challenges, this study introduces HSE-NET, a heterogeneous stacked ensemble model designed for sleep disorder classification. The proposed framework integrates Support Vector Machine, Logistic Regression, XGBoost, and Random Forest classifiers as base learners, while a neural network meta-learner combines their predictive outputs to produce the final decision. The experimental findings show that HSE-NET achieved an overall classification accuracy of 96%, with class-wise precision values ranging from 0.93 to 1.00 and recall values ranging from 0.87 to 1.00. These results indicate that the proposed approach performs better than individual machine learning models and recent ensemble-based methods. The reliability of the model was further supported through 5-fold cross-validation, while learning curve analysis showed stable convergence and strong generalization ability. By using physiological, behavioral, and demographic features, HSE-NET offers a scalable, interpretable, and cost-effective alternative for preliminary sleep disorder screening.