Auto-correlation Based Feature Extraction Approach for EEG Alcoholism Identification

Alcoholism severely affects brain functions. Most doctors and researchers utilized Electroencephalogram (EEG) signals to measure and record brain activities. The recorded EEG signals have non-linear and nonstationary attributes with very low amplitude. Consequently, it is very difficult and time-consuming for humans to interpret such signals. Therefore, with the significance of computerized approaches, the identification of normal and alcohol EEG signals has become very useful in the medical field. In the present work, a computer-aided diagnosis (CAD) system is recommended for characterization of normal vs alcoholic EEG signals with following tasks. First, dataset is segmented into several EEG signals. Second, the autocorrelation of each signal is computed to enhance the strength of EEG signals. Third, coefficients of autocorrelation with several lags are considered as features and verified statistically. At last, significant features are tested on twenty machine learning classifiers available in the WEKA platform by employing a 10-fold cross-validation strategy for the classification of normal vs alcoholic signals. The obtained results are effective and support the usefulness of autocorrelation coefficients as features.