Implementation of nonlinear classifiers for adaptive autoregressive EEG features classification

The objective of this work is to realize two nonlinear classifiers for the adaptive autoregressive Electroencephalography (EEG) features. The EEG features are modeled as adaptive autoregressive model and estimated using recurring least square algorithm. Nonlinear classification is performed using multilayer perceptron (MLP) and radial basal function neural network to classify extracted features for a two classes experiment. For validation, hands movement imaginations based experiments are conducted using low price EEG EPOC headset. A comparative study, carried out amongst the nonlinear classifiers and with a linear discriminant analysis, demonstrates the dominance of the MLP as nonlinear classifier.