Signal selection for sleep apnea classification

This paper presents a method for signals and features selection when classifying sleep apnea. This study uses a novel hierarchical parallel particle swarm optimization structure as proposed by the authors previously. In this structure, the swarms are separated into 'masters' and 'slaves' and access to global information is restricted according to their types. This method is used to classify sleep apneic events into apnea or hypopnea. In this study, ten different biosignals are used as the inputs for the system albeit with different features. The most important signals are subsequently determined based on their contribution to classification of the sleep apneic events. The classification method consists of three main parts which are: feature generation, signal selection, and data reduction based on PSO-SVM, and the final classifier. This study can be useful for selecting the best subset of input signals for classification of sleep apneic events, by attention to the trade of between more accuracy of higher number of input signals and more comfortable of less signals for the patient.