Automated Detection of Sleep Disorders Using Ensemble Learning on Clinical and Behavioral Features

Sleep disorders impact human health and their well-being which thereby necessitate an accurate and efficient diagnosing approach. Although traditionally, these disorders are diagnosed using clinical interviews, polysomnography test, etc., but they are time-consuming and costly. This paper focuses on the role of multiple machine-learning models in classifying different types of sleep disorders such as insomnia and sleep apnea. A comprehensive methodology is employed like data preprocessing, class balancing using SMOTE on the dataset containing clinical and behavioral attributes. which includes physical activity level, sleep duration, body mass index, stress level, blood pressure, and heart rate. Multiple models are examined, such as Logistic Regression, Decision Tree, Random Forest, Support Vector Machine, K-Nearest Neighbors, and Näive Bayes, alongside advanced ensemble methods like Voting Classifier, CatBoost, XGBoost, and HistGradient Boosting. Among all models, ensemble techniques, particularly the Voting Classifier and HistGradient Boosting, determined superior performance by obtaining an accuracy of 93.94%. The highest observed values of recall, precision, as well as F1-score are 0.94, computed by Voting and HistGradient Boosting, indicating strong model reliability across multiple classes. These results emphasize the importance of ensemble machine learning methods in building effective tools for automated sleep disorder diagnosis, contributing to more timely and accurate clinical assessments.