A Machine Learning Approach to Classify Sleep Stages of Rats

Identifying the vigilance states of the mammalian is an important research topic to bioscience in recently years, which the vigilance states is usually categorized as slow wave sleep, rapid eye movement sleep, and awake, etc. To discriminate difference vigilance states, a well-trained expert needs spend a long time to analyze a mass of physiological record data. In this paper, we proposed an automatic sleep stages classification system by analyzing rat’s EEG signal. The rat’s EEG signal is transferred by FFT and then extracted features. These extracted features are used as training patterns to further construct the proposed classification system. The proposed classification system contains two components, the principle component analysis (PCA) as the first component is used to projects the high dimensional features into lower dimensional subspace, and the k-nearest neighbor (k-NN) method as the second component is applied to identify the physiological state for a period of EEG signal. By experimenting on 810 periods of EEG signal, the proposed classification system achieves satisfactory classification accuracy of sleep stages. Key-Words: sleep stages, pattern classification, machine learning, PCA, k-NN, FFT