KMAN Publication Inc. (KMANPUB) Journal of Adolescent and Youth Psychological Studies (JAYPS) 2981-2526 7 Serial Number 48 2026 08 10 K-Nearest Neighbors Classification of Adolescents at Risk for Problematic Internet Use Based on Psychological and Behavioral Features K-Nearest Neighbors Classification of Adolescents at Risk for Problematic Internet Use Based on Psychological and Behavioral Features 1 19 EN Journal Article 2026 04 24 <table> <tbody> <tr> <td> <p><strong>Objective:</strong> This study aimed to classify adolescents at risk for problematic Internet use using a K-Nearest Neighbors algorithm based on a multidimensional set of psychological and behavioral features.</p> <p><strong>Methods and Materials:</strong> This cross-sectional predictive classification study was conducted with 704 adolescents aged 13–18 years enrolled in secondary schools in Ontario, Canada. Problematic Internet use was assessed using the Internet Addiction Test, and psychological predictors included depressive symptoms, anxiety, perceived stress, self-esteem, loneliness, and emotion regulation difficulties. Behavioral predictors included recreational Internet use, social-media use, online gaming, bedtime Internet use, smartphone checking, digital multitasking, sleep duration, physical activity, and interference of digital use with homework and face-to-face interactions. The sample was stratified into a training set of 528 participants and an independent test set of 176. Continuous predictors were standardized, and KNN hyperparameters were optimized through stratified 10-fold cross-validation. Model performance was evaluated using accuracy, sensitivity, specificity, precision, F1 score, balanced accuracy, and ROC-AUC.</p> <p><strong>Findings:</strong> Cross-validation identified a distance-weighted KNN model with k = 9 as optimal, yielding mean accuracy of 0.87, sensitivity of 0.82, specificity of 0.89, F1 score of 0.81, and ROC-AUC of 0.92. In the independent test sample, the model correctly classified 154 of 176 adolescents, corresponding to an accuracy of 0.88. Sensitivity was 0.83, specificity 0.89, precision 0.77, negative predictive value 0.92, F1 score 0.80, balanced accuracy 0.86, and ROC-AUC 0.92. Permutation analysis identified recreational Internet use, homework interference, emotion regulation difficulties, bedtime Internet use, depressive symptoms, and social-media use as the most influential predictors.</p> <p><strong>Conclusion:</strong> KNN classification based on combined psychological and behavioral features demonstrated strong discrimination of adolescents at risk for problematic Internet use and may provide a useful basis for early screening and targeted preventive assessment.</p> </td> </tr> </tbody> </table> https://journals.kmanpub.com/index.php/jayps/article/download/6223/11806