Key factors associated with social anxiety symptoms among Chinese adolescents: An exploration via machine learning models.

BACKGROUND: Numerous factors from the family environment, school contexts, and individual characteristics have been shown to be associated with adolescent social anxiety, but the relative importance of these factors is rather unclear. This study aimed to quantify the importance of these factors using machine learning techniques and identify key predictors. METHODS: Participants were 27,373 adolescents aged 10-18 years from a county in northern China. A total of 24 candidate variables were selected into three machine learning models (Random Forest [RF], Extreme Gradient Boosting [XGBoost], and Light Gradient Boosting Machine [LightGBM]). SHapley Additive exPlanations values were used to evaluate the relative importance of each predictor and to identify key features. Subgroup analyses were performed for gender, residence, number of siblings, and family economic status. RESULTS: The RF, XGBoost and LightGBM models identified 12, 13, and 12 key features, respectively. Eleven key features were selected from an integrative perspective, and nine were retained after considering subgroup analyses. These included verbal bullying victimization, self-esteem, school support, anxious attachment, avoidant attachment, social bullying victimization, age, sleep duration and resilience. CONCLUSION: Social anxiety symptoms among adolescents are associated with a range of factors across family, school, and individual domains. School contexts and individual characteristics show relatively higher importance for adolescent social anxiety, whereas family-related factors contribute less to it.