Property analysis of classifiers for sleep snoring detection

Abstract

Sleep snoring has become a serious concern for long term healthcare, as well as an indicative diagnosis to other critical diseases. Recently, many researches show that the snoring sound can be analyzed in different domain by various classifiers. In this paper, a comparison study with various classifiers has been provided for analyzing the advantages and the characteristics of the classifiers for sleep snoring detection. As the results, correlational filter multilayer perceptron neural network (f-MLP) and support vector machine (SVM) classifiers achieved the better generalization performances with the classification rate over 96% for the time domain snoring data set. Besides, the filtered data as the output of filter layer in f-MLP also provided a high discriminative feature set which made most of the classifier succeeded in their works. One important observation was that the ordinary multilayer perceptron (o-MLP) could not generalize with the time domain input data.