EEG Evaluation Method Using Nonlinear Modeling

An evaluation method for EEG (electroencephalogram) is described in this paper. The proposed method uses a kind of non-linear autoregressive model by use of MLP (multilayered perceptron). The MLPs are trained by observed signals for short term periods. When the training is completed, each MLP can be thought as a model of EEG generator. To evaluate the change of the generator, variation of connection weights in the MLPs is obtained by the principal component analysis. Advantage of this method is that few prior information is necessary in comparison with conventional frequency analysis.