["Developmental Science, Volume 29, Issue 4, July 2026. ", "\nABSTRACT\n\nCyberbullying poses a substantial threat to adolescents’ well‐being, yet prevention efforts remain limited by insufficient understanding of its multilevel determinants. Guided by ecological systems theory, this study applies explainable machine learning (ML) to examine factors associated with cyberbullying perpetration across individual, family, peer, class, school, and online contexts. Questionnaire data from 2286 adolescents (Mage = 13.46 years, SD = 0.93; 11–16 years) were analyzed. Random Forest and XGBoost achieved out‐of‐sample accuracies of 87.35% and 85.95%, respectively. Model‐based importance analyses consistently highlighted Childhood Psychological Abuse, Adverse Peer Interactions, and Cyberbullying Victimization as the highest‐ranked predictors. At the system level, variables from the Family, Individual and Cyber contexts accounted for a substantial share of model importance, indicating their salience for intervention design. These findings prioritize psychosocial targets for prevention and demonstrate how explainable ML can synthesize questionnaire data to inform multi‐tiered strategies against adolescent cyberbullying.\n\n\nSummary\n\nBy adopting an ecosystem‐based perspective, this study utilizes comprehensive, multi‐level datasets to model adolescent cyberbullying.\nExplainable machine learning techniques are employed to systematically identify and interpret key predictors of cyberbullying perpetration.\nChildhood psychological abuse exhibits strong predictive power for adolescent cyberbullying perpetration.\nThe study proposes a methodological framework for applying explainable machine learning to structured questionnaire data in psychological research.\n\n\n"]