Applying Variable Ranking to Oximetric Recordings in Sleep Apnea Diagnosis

Abstract

This study is aimed at assessing the usefulness of variable ranking techniques for feature selection in the context of sleep apnea hypopnea syndrome (SAHS) diagnosis from blood oxygen saturation (SpO2) recordings. Time, frequency, linear, and nonlinear analyses were carried out to compose an initial feature set from oximetry. Principal component analysis (PCA) and fast correlation-based filter (FCBF) were used to derive suitable feature subsets. Support vector machines (SVMs) were applied in the classification stage. A total of 240 subjects suspected of suffering from SAHS composed the population under study. FCBF-based feature subsets significantly outperformed PCA in the test set. A SVM with 5 input features from FCBF achieved the highest performance: 86.5% sensitivity, 83.3% specificity, and 85.4% accuracy. Our results suggest that a suitable analysis of the feature space by means of variable ranking techniques could provide useful information to assist in SAHS diagnosis.