<![CDATA[Smartphone addiction among adolescents threatens mental health, sleep, academic performance, and social functioning, creating an urgent need for objective, scalable screening tools. This study evaluates whether a weighted RBFNN (Radial Basis Function Neural Network) with learnable per-feature weights can improve early identification of smartphone addiction in adolescents. Using a questionnaire-derived dataset of 394 participants, features were standardized, hidden-unit centers were initialized with k-means clustering, and both conventional and weighted RBFNN architectures were trained and compared under stratified ten-fold cross-validation while sweeping the number of hidden units. Models were assessed by accuracy, precision, recall, F-measure, area under the receiver operating characteristic curve, and computation time. The weighted RBFNN consistently outperformed the conventional variant and previously reported baselines, achieving a mean cross-validation accuracy 97.99% across all cluster settings, with best performance of 98.48% (at cluster 6) and an area under the curve near 0.989, with very low false positive and false negative rates and reduced computation time at larger cluster settings. Learnable feature weighting mitigated noisy predictors and improved generalization, while results underscored sensitivity to cluster-count selection and preprocessing choices. These findings indicate that a weighted RBFNN is a promising, efficient approach for automated adolescent smartphone addiction screening.]]>
Copyrights © 2026
