Automatic Sleep Stage Classification by a Density - Distance- Based K - means Clustering Algorithm with Amendments

Abstract

Sleep staging has important significance for the monitoring, prevention and treatment of sleep disorders. In this study, the clustering algorithm of K-means is investigated in order to realize the automatic sleep stage classification. Unlike the traditional K-means clustering algorithm, density-distance-based processing procedures were presented to determine the cluster centers. Additionally, the clustering results were amended based on the regularity of sleep stages during one's overnight sleep process. The presented algorithm was tested and analyzed in detail with six subjects from the Sleep-EDF database. The averaged accuracy is about 74%. Comparing with the traditional K-means clustering algorithm, the developed algorithm achieved more accurate and reasonable sleep staging results which can be an assistant tool for clinical inspection.