Selection of features for detection of Obstructive Sleep Apnea events

Generalization performance of a classifier depends primarily, on the selection of good features, i.e., features that represent maximum separation between the classes. So, the optimum number of features are to be identified that maximizes the classifier performance. In the present paper, two different feature selection methods i.e. F-score and Support Vector Machine based Recursive Feature Elimination technique (SVMRFE) are evaluated to find the optimum set of features extracted from short segment overlapping windows of respiration signal, which will help to identify the abnormal breathing events occur during sleep. The extracted features are from time domain, frequency domain and non-linear analysis of the window wise segmented respiration signal. The SVM-RFE based feature selection method technique found to be better in comparison with F-score based feature selection method. The present evaluation is based on the study of total 14 records of Obstructive Sleep Apnea (OSA) subjects collected from MIT-BIH database.