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

Background: 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). Methods: 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. Results: 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. Conclusion: 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.