Unveiling Roots of Chinese Adolescent Cyberbullying Through Explainable Machine Learning Approach

Area of Science:

• Psychology
• Computer Science
• Public Health

Background:

• Cyberbullying is a major threat to adolescent well-being.
• Limited understanding of multilevel determinants hinders effective cyberbullying prevention.
• Ecological systems theory provides a framework for examining environmental influences.

Purpose of the Study:

• To apply explainable machine learning (ML) to identify multilevel determinants of cyberbullying perpetration in adolescents.
• To analyze factors across individual, family, peer, class, school, and online contexts.
• To inform multi-tiered intervention strategies for adolescent cyberbullying prevention.

Main Methods:

• Utilized questionnaire data from 2286 adolescents (ages 11-16).
• Employed explainable ML algorithms (Random Forest, XGBoost) for predictive modeling.
• Conducted model-based importance analyses to rank predictor significance.

Main Results:

• Random Forest and XGBoost achieved high predictive accuracies (87.35% and 85.95%).
• Childhood Psychological Abuse, Adverse Peer Interactions, and Cyberbullying Victimization were top predictors.
• Family, Individual, and Cyber contexts significantly contributed to model importance.

Conclusions:

• Explainable ML effectively synthesizes complex questionnaire data for cyberbullying research.
• Childhood psychological abuse is a critical target for intervention.
• Findings support the development of multi-tiered, ecosystem-informed prevention strategies against adolescent cyberbullying.