ASSESSMENT OF FEATURE SELECTION AND CLASSIFICATION APPROACHES TO ENHANCE INFORMATION FROM OVERNIGHT OXIMETRY IN THE CONTEXT OF APNEA DIAGNOSIS

摘要: This study is aimed at assessing the usefulness of different feature selection and classification methodologies in context sleep apnea hypopnea syndrome (SAHS) detection. Feature extraction, stages were applied to analyze blood oxygen saturation (SaO2) recordings order simplify polysomnography (PSG), gold standard diagnostic methodology for SAHS. Statistical, spectral nonlinear measures computed compose initial set. Principal component analysis (PCA), forward stepwise (FSFS) genetic algorithms (GAs) select subsets. Fisher's linear discriminant (FLD), logistic regression (LR) support vector machines (SVMs) stage. Optimum from each combination these approaches prospectively validated on datasets two independent units. FSFS + LR achieved highest performance using a small subset (4 features), reaching 83.2% accuracy validation set 88.7% test Similarly, GAs SVM also high generalization capability number input features (7 with 84.2% 84.5% Our results suggest that reduced subsets complementary (25% 50% total features) classifiers ability could provide high-performance screening tools