Improving Sleep Disorder Diagnosis Through Optimized Machine Learning Approaches

Abstract

The accurate diagnosis of sleep disorders such as obstructive sleep apnea and insomnia is extremely significant because of the significant health impacts that can occur. This is the research where we try to understand best ML approaches to predict sleep problems, using the Sleep Health and Lifestyle dataset available in Kaggle. ANOVA is used for feature selection and then resampling is performed using SMOTEENN for class imbalance . The seven classifiers that are used to derive the new engineered features are: RF, Gradient Boosting, Gaussian Naive Bayes, KNN, DT, LR and SVM. There are seven new features added in all. We use different ML techniques such as LR, RF, DT, SVM, ET, Extreme Gradient Boosting, Gradient Boosting, LightGBM, CatBoost, AdaBoost, Naive Bayes, Voting Classifier and Stacking Classifier. The designed feature Voting Classifier has good prediction performance, with a 97.3% accuracy for both 5 and 2 feature sets. With original features, Stacking Classifier gave 88.0% accuracy while with two features it gave 80.0% accuracy. To facilitate clinical decision assistance, explainable AI methods such as LIME and SHAP are integrated to interpret model predictions, hence boosting transparency.