Presenting efficient features for automatic CAP detection in sleep EEG signals

Research findings show that several diseases can be detected by quantitative analysis of sleep signals. Detecting and analyzing cyclic alternative pattern (CAP) is an essential part of the sleep analysis. Although some methods have been suggested for automatic CAP detection, none of them can provide an acceptable accuracy. In this paper, a family of entropy based features is evaluated by support vector machine (SVM), K-nearest neighbor (KNN) and linear discriminant analysis (LDA) to distinguish CAP from non-CAP parts. To assess the suggested feature set, sleep EEG of 4 healthy subjects and 4 patients are analyzed by the conventional and the suggested features. Comparative results demonstrate that a subset of suggested features can drastically outperform the previous features for both groups of healthy and patients.