An investigation of data mining based Automatic Sleep Stage Classification techniques

Abstract

Sleep quality is highly significant for the people's overall health. A standard diagnosis for sleep-related syndromes and illnesses is Polysomnography (PSG) or a sleep test in a controlled laboratory. However, PSG requires a sleep specialist to interpret bio-signals collected. It is a time consuming procedure. One of the fundamental step in the PSG is Sleep Stage Classification (SSC). In this study, we propose an investigation of Automatic Sleep Stage Classification (ASSC) using data mining techniques as an alternative to the PSG in order to reduce the time necessary for accurately diagnosing sleep quality. We studied 2,535 subjects' polysomnographic data with 14 channels of biomedical signals from the Sleep Heart Health Study (SHHS) Dataset. Subsequently, four data mining techniques including Decision Trees, Random Forests, Neural Network, and k-Nearest Neighbors were selected to compare the classification performances. The classification results in k-Nearest Neighbors achieved the greatest accuracy at 83.76%.