An approach for classifying alcoholic and non-alcoholic persons based on time domain features extracted from EEG signals

Automatic detection of alcoholics can play important role in reducing the number of road accidents and social violence. In this paper, a scheme for classifying alcoholic and non-alcoholic subjects is developed based on autocorrelation domain feature extraction. In order to investigate the difference in characteristics of EEG signal recorded after exposing visual stimulus to alcoholic and non-alcoholic subjects, instead of using conventional narrow band filtering, high pass IIR filterwith zero phase distortion is used, which preserves Gamma band and all higher frequencies. Next the reflection coefficients of the filtered EEG signal are extracted by using the autocorrelation values in a recursive fashion. Instead of AR parameters, first few reflection coefficients are proposed as feature, which provide better feature consistency and noise immunity with reasonable computational burden. For the purpose of classification, K nearest neighbor (KNN) classifier is employed in leave one out cross validation technique. From extensive simulation on a publicly available EEG dataset, it is found that the proposed scheme can classify alcoholic and non-alcoholic subjects with a very highlevel of accuracy in comparison to some of the existing methods.