Association of problematic short-video use with adolescent depression: a cross-sectional study and machine learning identification model construction

BACKGROUND: Problematic short-video use (PSVU) is prevalent among adolescents and is increasingly recognized for its association with adverse mental health outcomes. This cross-sectional study aims to explore the association between PSVU and depression using SFPSVUS, and to develop an interpretable machine learning (ML) model for identifying adolescent depression. METHODS: A cross-sectional analysis was conducted on data from 4000 adolescent participants. Multivariate Logistic Regression and restricted cubic spline (RCS) methods were used to assess the association between PSVU and depression. Feature variables were identified via Boruta feature selection and validated using Shapley Additive Interpretation (SHAP) methods. Nine ML models, including XGBoost, support vector machine (SVM), and random forest (RF), were developed to identify adolescents with depression. RESULTS: A total of 3816 adolescents were included. The SFPSVUS demonstrated acceptable reliability (Cronbach’s α = 0.70) in the current sample. Multivariate Logistic Regression analysis revealed a significant positive correlation between PSVU and depression. RCS analysis indicated a significant nonlinear association between PSVU and adolescent depression. Six variables selected via Boruta feature selection were used to develop nine ML models, with the RF model demonstrating superior performance in identifying depression (AUC = 0.863) in the test set. PSVU emerged as an important feature in the model, alongside emotion regulation, non-suicidal self-injury (NSSI) and sleep quality. CONCLUSIONS: PSVU shows a significant positive association with adolescent depression and serves as an important feature in ML identification models. CLINICAL TRIAL NUMBER: Not applicable.