Advanced measure selection in automatic NREM discrimination based on EEG

Smart homes have been proposed for senior citizens aiming to improve their quality of life. In order to monitor the seniors ' sleep condition in smart homes to protect them from sleep disorder, an automatic sleep staging system is necessary. Furthermore, measure selection is a crucial step in an automatic sleep staging system. In this paper, we present three advanced combination schemes among 15 sleep measures, which have been most commonly used in the automatic discrimination of sleep stages. For testing the validity of these combination schemes, we apply each combination of measures into a BP network to calculate the discrimination rates of NREM. The result of experiment shows that the combination method of two different ideologies of nonlinear measures is better than those including only one of them. And combining measures that have high accuracy rate in discriminating three sleep stages respectively can get effective results. Finally, we can get an optimum measure combination, composed of measures including EDGE, KK, delta, SaEn and LLE. By testing, the total accuracy rate of sleep staging reaches 85.23%, 69.57%, 86.11% and 100% in SI, S2 and SWS respectively. This result is desirable under the condition of only using EEG.