Resting State EEG Based Classification of Premature Internet Addicts

Abstract

The proposed work exploits the ability of machine learning classification model in diagnosis of Internet Addiction using resting state Electroencephalogram signals (rs-EEG). Young’s Internet Addiction Test (IAT) interpretation is used as the ground truth for training the model. Time domain features of rs-EEG are used as predictors for IA. The problem is cast as a classification problem, one using Eyes Closed (EC) and another using Eyes Open (EO) rs-EEG signals. The study sample consists of 29 premature addicts and 30 non addicts from the publicly available Mind-Brain- Body (LEMON) dataset. Different classification models viz., Support Vector Machine, Logistic Regression, k-Nearest Neighbour, Random Forest and Balanced Bagging Trees are evaluated for identifying optimal performing model for EC and EO signals. The result shows an optimal accuracy of 95% and 94% for EC and EO based classification models respectively. The present work contributes towards neurophysiological diagnosis of IA. It also provides quantitative evidence for effects of IA. The work is relevant for mental health professionals and data scientists in health care domain.