Alcohol use disorder (AUD) causes systemic damage to the human body, including brain structure and function. Computer-aided AUD diagnosis offers fast and reliable detection of AUD to prevent further harm from alcohol consumption. Various machine learning models and feature engineering techniques have been proposed for high system accuracy, reliability, and interpretability. EEG features from brain connectivity are promising due to their interpretability properties. This paper aims to propose a novel feature from task-based EEG signal data for computer-aided AUD diagnosis. The functional connectivity was calculated using the phase lag index (PLI) by choosing eight EEG channels from brain areas processing visual stimuli. The resulting connectivity values were evaluated using discriminant analysis. The proposed feature has yielded a significant discriminant function that proved its differentiation properties. The highest differentiation properties resulted from the gamma band discriminant function with p-value < 0.001 and a canonical correlation of 0.823. The classification accuracy reached 91.7%, and the leave-one-out cross-validation accuracy of 89.6%, showing consistent generalization.
