Alcohol use Disorder Detection and Classification from EEG Signals Using 2D Wavelet Transform, Machine Learning and Deep Learning

Abstract

Alcohol Use Disorder (AUD) is a psychological condition in which individuals lose control over alcohol consumption despite awareness of its harmful consequences. Electroencephalogram (EEG) signals offer a noninvasive means to characterize brain activity and have shown potential for distinguishing alcoholic and non-alcoholic subjects. However, highly accurate and automated detection methods for AUD remain limited. This study aims to improve classification performance using multichannel EEG signals transformed through time–frequency analysis combined with machine learning (ML) and deep learning (DL) techniques. Sixty-four–channel EEG recordings were converted into images and processed using a three-level Discrete Wavelet Transform (DWT). Statistical features were extracted from each wavelet sub-band and classified using various ML and DL models. Experimental results on real EEG data demonstrate that the Random Forest classifier achieves the highest accuracy of 93.77%, outperforming other methods. Comparative analysis indicates competitive performance relative to prior studies. These findings highlight the effectiveness of 2D DWT for EEG-based AUD detection and suggest that further improvements may be achieved through alternative mother wavelet selection, decomposition levels, and advanced classification techniques