MREM-IUP - A MultiRegressor Based Ensemble Model for Assessing the Internet Addiction in Youth Using Physical and Behavioural Indicators

Abstract

The rising prevalence of internet use among adolescents has led to increased research on Problematic Internet Usage (PUI), a behavioral issue linked to negative impacts on mental and physical health. Existing assessment tools for PUI often lack precision, overlooking key factors like physical activity levels. This study proposes a Multi-Regressor Ensemble Model for Internet Usage Prediction (MREM-IUP) to predict PUI severity in adolescents by integrating physical activity data, demographics, and behavioral assessments. Using a rich dataset from the HealthyBrain Network, our model combines advanced machine learning algorithms—LightGBM, XGBoost, CatBoost, and TabNet—within a Voting Regressor to predict the Severity Impairment Index (SII). Our model achieved an optimized Quadratic Weighted Kappa (QWK) score of 0.92, indicating high accuracy in predicting PUI severity. Additionally, significant correlations were found between low physical activity levels and higher PUI scores, highlighting physical fitness as a potential protective factor. The proposed model offers a novel, data-driven approach to assessing PUI, with potential applications in developing targeted interventions that promote healthier online habits among youth.