A Novel Machine Learning Approach for High-Performance Diagnosis of Premature Internet Addiction Using the Unfolded EEG Spectra

Abstract

Due to the increasing presence of internet-based applications in our private and professional environments, internet addiction (IA) has emerged as a universal issue today’s society. Clinical diagnosis of IA is still in its beginnings, resulting in delays or failures in psychological interventions. Predictive healthcare analytics can increase the speed and robustness of diagnosis, resulting in faster preventive interventions. By unfolding the conventional frequency bandwidths into a fine-graded equidistant 88-point spectrum, we present a method for detecting premature IA. We identified the most predictive frequency sub-bands that differenti-ate healthy persons from people suffering from preliminary stages of IA (10.5-11 Hz, 21.5-22 Hz, 22.5-23 Hz). With a balanced accuracy of 94.17%, our results set a new benchmark in detecting IA using resting-state electroencephalography recordings. We provide evidence that IA can lead to structural changes in the brain and thus indicates the need for it to be considered alongside other generally recognized disorders.