The impact of problematic social media use on adolescent subjective well-being: insights from machine learning

Higher subjective well - being (SWB) among adolescents predicts higher educational achievements, work capabilities, and job satisfaction. Problematic social media use may adversely affect adolescents' SWB. This study aims to identify key predictors of adolescents' SWB by combining machine learning with multiple logistic regression methods, and to examine the relationship between problematic social media use and SWB. This study was based on data from the 2017/2018 Health Behaviour in School-aged Children (HBSC) survey. The data included the situations of problematic social media use of 187,090 adolescents aged between 10 and 16. The XGBoost and multivariate logistic regression models were used to explore the impacts of problematic social media use on adolescents' SWB. With the exception of the time reduction failure dimension, all other dimensions of problematic social media use were negatively associated with SWB. Gain and SHapley Additive exPlanations (SHAP) analyses indicate that feeling escape, age, sex, and family conflict are key predictors of SWB. They also suggest that there may be complex interaction patterns among these variables. Specifically, feeling escape and family conflict account for 49.26% of the total variance, while age and sex account for 25.5%. Multivariate logistic regression results indicate that feeling escape, family conflict, increasing age, and female gender are all significantly associated with lower SWB. Therefore, future interventions should go beyond simply limiting social media usage time and focus on adolescents' emotional regulation skills, the quality of family communication, and tailored support for older adolescents and girls.