Development and interpretability analysis of a risk prediction model for Problematic Internet Use in adolescents

Problematic Internet Use (PIU) is a significant mental health concern in adolescents. This study aimed to develop machine learning (ML)-based predictive models to identify and explain the key risk factors for PIU. A total of 8176 junior high school students from Henan Province were surveyed from April to May 2023. The dataset was randomly divided into training and test sets in an 8:2 ratio. Five ML algorithms (Logistic Regression (LR), Random Forest (RF), Extreme Gradient Boosting (XGBoost), Multilayer Perceptron (MLP), and LightGBM) were used to predict PIU, and feature importance was determined using SHapley Additive exPlanations (SHAP). The XGBoost model, which achieved the highest area under the curve (AUC), was selected for detailed analysis and individualized prediction explanations. The five most important predictors of PIU were negative life events, school connectedness, self-esteem, parent-adolescent cohesion, and psychological resilience. Importantly, SHAP interaction analysis suggests that the relationship between resilience and PIU risk may vary depending on self-esteem, with the influence of low resilience varying across different levels of self-esteem. The results of this study indicate that ML models combined with SHAP may be useful in predicting PIU and exploring its psychosocial determinants. These findings imply that self-esteem and psychological resilience could play an important role, offering preliminary insights for clinicians and educators in addressing PIU among adolescents.