Predicting and structuring adolescent academic burnout: A multi-ecological framework via machine learning and network analysis

Academic burnout is a prevalent concern with significant implications for adolescent development, whereas variable-centered approaches often provide limited insight into its multi-ecological heterogeneity. This study combines machine learning (ML) with network analysis to examine the multi-level ecological structure of academic burnout. Using a cross-sectional sample of 4870 Chinese adolescents, seven ML models were compared, with the Support Vector Machine (SVM) achieving the best classification performance (AUC = 0.830). SHapley Additive exPlanations (SHAP) highlighted non-linear associations among predictors, with depressive symptoms, digital stressors (e.g., social media online vigilance), and resilience showing high relative importance. Network analysis showed that depressive symptoms occupied a central position across ecological domains, digital stressors showed cross-contextual connectivity, and teacher-student relationships exhibited a bridging position between school context and individual burnout. These findings suggest that addressing academic burnout may benefit from moving beyond broad support toward more targeted, system-informed approaches that consider core structural nodes and cross-domain.