Analysis and classification of EEG signals using mixture of features and committee neural network

Abstract

Electroencephalography signal is the recording of electrical activity of brain, provides valuable information of the brain function and neurological disorder. this paper proposed committee neural network for classification of EEG signals. Committee neural network consists of different neural network that used multilayer perceptron back propagation algorithm. The number of input node and hidden node selection for artificial neural network remains an important issues, as over parametrized ANN gets trapped in local minima resulting non convergence of ANN structure during training. Redundant features and excessive hidden
nodes of ANN increases modeling complexity without improving discrimination performance. Therefore optimum design of neural network which intern optimizes the committee neural network is required towards real time detection of EEG signals. The present work attempts to: (i) develop feature extraction algorithm which combines the score generated from autoregressive based feature and wavelet based feature for better classification of EEG signals, (ii) a two-level committee neural network is proposed based on the decision of several neural networks, (iii) select a set of input features that are effective for identification of EEG signal using genetic algorithm, (iv) make certain optimum selection of nodes in the hidden layer using genetic algorithm for each ANN structure of two-level CNN to get effective
classification of EEG signal. It is observed that the performance of proposed technique is better than the earlier established techniques (combined neural network based model and wavelet/ mixture of experts network based approach) and the technique that uses artificial neural network with back propagation multilayer perceptron