Statistická analýza polysomnografických dat pro detekci spánkových poruch

This master’s thesis analyzes predictors of sleep disorders using data from the Sleep Heart Health Study, comparing underlying mechanisms in younger and older populations. Due to the high multicollinearity of medical data, Principal Component Analysis (PCA) was utilized, followed by modeling hidden dependencies using Bayesian Networks. The results revealed fundamental age-related differences. In the younger cohort, logistic regression achieved good predictive performance (AUC > 0,73), identifying medication and a history of sleep apnea as key risks. In seniors, while the inferential logistic regression model highlighted the strong impact of psychosocial factors (such as marital status), linear models failed for classification purposes due to data non-linearity. Successful prediction for this group was achieved only by applying the Naive Bayes algorithm to orthogonal PCA components. The thesis demonstrates the necessity of combining dimensionality reduction with machine learning algorithms when processing complex biomedical data.