Neural, Fuzzy And Neurofuzzy Approach To Classification Of Normal And Alcoholic Electroencephalograms

According to the literature, many psychiatric phenotypes, brain disorders and/or mental tasks can be detected by analyzing EEG signals. One such a psychiatric phenotype is alcoholism. In this paper the parameters of second order autoregressive model, peak amplitude of the power spectrum, mean of absolute value and the variance of the signal are extracted as features of the signal. The dimension of the feature vector is then reduced by means of PCA. Next a method based on fuzzy inference system as a fuzzy approach in classification is investigated. In this method first the data in each class is divided into two clusters separately and a Gaussian membership function is defined for each cluster. Classification is performed by means of if-then rules generated in the previous step. Then an adaptive neurofuzzy inference system is used for classification. Due to the ability of the neurofuzzy inference system to be trained higher classification accuracy is achieved. Finally with the use of a multilayer perceptron structure it is shown that an accuracy of 100% can be achieved for separating the two classes.