Computational Phenotyping of Sleep Disorder Using Machine Learning

Early identification of conditions such as insomnia, sleep apnea, and hypersomnia remains difficult because conventional diagnosis often relies on subjective evaluations and limited access to polysomnography. To address this gap, this work introduces a machine-learning-based computational phenotyping approach for automatically identifying sleep-disorder patterns. The dataset, which included demographic details, lifestyle habits, physiological metrics, and sleep-quality measures, was carefully preprocessed using normalization, outlier handling, and imputation strategies. Key variables affecting sleep health were selected through correlation analysis and recursive feature elimination. Several supervised algorithms—including Logistic Regression, Support Vector Machines, Random Forest, and Gradient Boosting—were developed and compared. Among them, the Random Forest classifier delivered the best performance with an accuracy of 92.4%, indicating strong predictive capability for sleep-disorder categorization. The findings demonstrate that computational phenotyping can serve as an efficient, low-cost complement to traditional sleep-lab assessments and holds promise for personalized digital health applications.