Detection of Alcoholism from EEG signals using Spectral and Tsallis Entropy with SVM

Abstract

Alcoholism is the most common disorder. It leads to various emotional, behavioral, and cognitive brain defects. Finding and extracting discriminative biological markers, which are correlated to healthy brain patterns and alcoholic brain pattern helps us to utilize automatic methods for detecting and classifying alcoholism. In this study, we evaluate the complexity of Electroencephalogram (EEG) dynamics using Spectral entropy (SpEn) and Tsallis entropy (TsEn) and results are used to distinguish alcoholic from Controlled subjects. Our analysis shows that the SpEn and TsEn values for alcoholic EEG signals and lower compared to Controlled. The feature vectors from SpEn and TsEn are provided to various supervised and unsupervised classification techniques with 10-fold cross-validation. Among these Support Vector Machine (SVM) and K-Nearest Neighbors (KNN) gave maximum accuracy of 95.6% and 92.2% respectively. Hence, our hybrid approach of combining SpEn and TsEn helps to diagnose alcoholism to medical practitioners.