Automatic EEG Signal Classification Based On Machine Learning

Electroencephalography (EEG) is the electrical activity of human brain. It can reflect the physiological functions and mental states of the brain. So EEG has been widely used in the diagnosis of neurological diseases, Brain-Computer interface, and so on. With the development of digital signal processing technology and computer science, EEG acquisition equipments have been improved a lot, and people have had a deeper understanding about EEG. Therefore, traditional EEG analysis hasn't been able to meet people's needs, and automatic EEG signal classification based on machine learning has been sought by many researchers. EEG signal classification based on machine learning consists of three parts:EEG feature extraction, feature transform, and classification.EEG feature extraction plays an important role in such a problem. This paper proposes a new EEG feature extraction based on echo state network (FE-ESN). Unlike most of the traditional EEG feature extraction methods, FE-ESN is not to find some describing quantities of EEG signals, but to identify the systems which generated the EEG signals. So it's an unsupervised EEG feature extraction method, and the information lost in the process of feature extraction is very small. Automatic EEG signal classification based on such an EEG feature extraction method can reach high accuracy.Traditional automatic EEG signal classification methods usually include two parts:EEG feature extraction and pattern classification.In this paper, principle component analysis (PCA) and linear discriminant analysis (LDA) are employed to reduce the dimensionality of feature space. Results show that a three-step automatic EEG signal classification method (EEG feature extraction, dimensionality reduction of feature space, and classification) is more scientific and rational.Relevance rector machine (RVM) and extreme learning machine (ELM) are two recently proposed machine learning methods. The both are promising to improve the performance of automatic EEG signal classification system. This paper studies some automatic EEG signal classification systems based on the two pattern recognition methods. Results show that both of them can well component the tasks of automatic EEG signal classification for epilepsy.