A Machine Learning Approach for EEG Brain Signal Classification

The analysis of signals from electroencephalography is an important aid for the patient condition diagnostics. A thorough analysis of the EEG allows you to obtain valuable information and improve understanding of the mechanisms that cause various kinds of disorders, such as epilepsy. In present paper, the construction of an EEG signal classifier based on an artificial neural network is considered. This article compares various types of classifiers, including those based on a complex-valued neural network. An open database is utilized as source data. Since brain signals have non-stationary nature, preprocessing is carried out using the theory of wavelet analysis. A discrete-wavelet transform is performed, the coefficients of which are fed to the input of the neural network. A various types of the wavelet families, including wavelet atomic functions are applied. Some wavelet atomic functions have promising spectral characteristics that motivates to apply such a functions in biomedical problems as well.