Investigation of New Statistical Features for BCI Based Sleep Stages Recognition through EEG Bio-signals

Abstract

Abstract. Electroencephalogram (EEG) is one of the oldest techniques available to read brain data. It is a methodology to measure and record the electrical activity of brain using sensitive sensors attached to the scalp. Brain's electrical activity is visualized on computers in form of signals through BCI tools. It is also possible to convert these signals into digital commands to provide human-computer interaction (HCI) through adaptive user interfaces. In this study, a set of statistical features: mean entropy, skew-ness, kurtosis and mean power of wavelets are proposed to enhance human sleep stages recognition through EEG signal. Additionally, an adaptive user interface for vigilance level recognition is introduced. One-way ANOVA test is employed for feature selection. EEG signals are decomposed into frequency sub-bands using discrete wavelet transform and selected statistical features are employed in SVM for recognition of human sleep stages: stage 1, stage 3, stage REM and stage AWAKE. According to experimental results, proposed statistical features have a significant discrimination rate for true classification of sleep stages with linear SVM. The accuracy of linear SVM reaches to 93% in stage 1, 82% in stage 3, 73% in stage REM and 96% in stage AWAKE with proposed statistical features.