Figure 2.

Schematic workflow. All derived data were first subjected to mean comparison for the three disease groups. Next, all data were divided into training and testing datasets. Five matching learning approaches were employed in this study, namely, a logistic regression (LR) model, k-nearest neighbors (kNN), naive Bayes (NB), random forest (RF), and support vector machine (SVM). Two types of models were established by using these five approaches, including the oximetry model, which was only trained using oximetry parameters, and combined models, which were trained by both oximetry and anthropometric parameters. The trained model with the highest accuracy in the training outcomes for both model types was used to predict the testing dataset. Abbreviations: ID: insomnia disorder; ArTH: respiratory arousal threshold; OSA: obstructive sleep apnea.