Classifying Internet Addiction Using Machine Learning Approach: A Study Among Adolescents in Bangladesh

TL;DR

This study uses machine learning to identify risk factors for internet addiction among Bangladeshi adolescents, finding that depression, loneliness, and internet use time are key predictors.

Contribution

The study introduces a machine learning framework to classify internet addiction in Bangladeshi adolescents, identifying novel predictive features and achieving high classification accuracy.

Findings

30.1% of 385 adolescents reported internet addiction.

SVM linear kernel model achieved 81.9% accuracy in classifying IA.

Depression, loneliness, and internet use time were significant predictors of IA.

Abstract

Internet addiction (IA) among adolescents is growing worldwide. Online temptation is particularly strong for adolescents due to rapid physical and cognitive development. IA may impair their mental, emotional, social, and physical health. Few traditional studies were conducted in Bangladesh. Thus, this study aimed to identify adolescents’ IA risk factors using advanced machine learning (ML).

A total of 385 individuals were convenience sampled and surveyed using the Patient Health Questionnaire‐9 (PHQ‐9), the UCLA Loneliness Scale (UCLA‐3), and Young's IA Test (IAT‐20) to measure the prevalence of depression, loneliness, and IA. Boruta found IA prevalence classifying factors. We evaluated decision tree (DT), support vector machine (SVM), logistic regression (LR), and random forest (RF) classification models using confusion matrix, receiver operating characteristic (ROC) curves, and k‐fold cross‐validation.

Among 385 respondents, one‐third (30.1%) reported IA. Participants’ fathers’ education, favorite activity, loneliness, smoking status, depression, and internet use time were selected as important features classifying IA. The performance was tested on the basis of five different classification techniques overall: the SVM linear kernel model (accuracy = 0.819, specificity = 0.869, sensitivity = 0.687, precision = 0.666, area under the ROC curve [AUC] = 0.890, k‐fold accuracy = 0.801) performed better and authentically classified IA.

Raising awareness among adolescents and their parents is crucial because IA is frequent. The ML framework can identify significant prognostic indicators and classify this IA problem more accurately, helping policymakers, stakeholders, and families understand and prevent this crisis by improving policy‐making strategies and counseling services.

The infographic outlines a study on classifying internet addiction (IA) among adolescents in Bangladesh using machine learning. The study involved 385 participants aged 13–19, focusing on key predictors such as depression, loneliness, smoking status, and internet use time. Data were collected using the IAT‐20, PHQ‐9, and UCLA‐3 scales. The Boruta algorithm was used with models like decision tree, logistic regression, and random forest, achieving an accuracy of 81.9%, sensitivity of 68.7%, specificity of 86.3%, and an AUC of 0.89. The findings highlight significant prevalence rates for internet addiction, depression, and high loneliness among participants.

Figures