Individual differences in adolescents’ vulnerability to short video addiction: a SHAP-based typology and risk index approach
Abstract
Adolescents differ substantially in their vulnerability to short video addiction, yet early-stage risk often remains covert and difficult to detect. Existing research has been limited in capturing the complex, multidimensional nature of these individual differences, particularly in identifying key predictive factors and modeling heterogeneous risk patterns. Consequently, important gaps remain in understanding how diverse psychological and environmental factors jointly contribute to early addiction risk. Data were collected from 1276 Chinese adolescents (703 boys, 573 girls; M_age = 14.66, SD = 2.33) in this study. A supervised machine learning framework was applied to predict short video addiction risk. Multiple algorithms were compared, SHAP analysis was used to identify key predictors and construct a composite risk index, and person-centered classification was conducted to detect distinct risk profiles. Results showed that LightGBM achieved the best predictive performance. Seven key predictors were selected from nearly 30 indicators to construct a SHAP-based risk index, and K-means clustering identified three distinct addiction profiles. This study advances theory by clarifying mechanisms underlying early risk of adolescent short video addiction and addressing prior gaps in early detection, while offering practical value through a validated risk identification tool and tailored intervention strategies for distinct addiction profiles.