On the EEG-based Automated Detection of Alcohol Dependence

Power and magnitude square coherence estimates evaluated for EEG of alcoholics and control participants were used to attempt an automated discrimination of individuals suffering alcohol dependence. The estimates were obtained for non-overlapping consecutive EEG fragments of 0.5 second duration with parametric analyzers and used as features for Euclidean, Fisher, and Regression-based classifiers. Implementing the leave-one-out cross- validation technique, the highest unbiased classification accuracy, sensitivity of 66.45% and selectivity of 67.12%, was observed from the Regression-based classifier when θ-rhythm power estimates for all EEG electrodes were used as classification features.