Key Predictors of Bullying: Global Trends from 2001–2014 via Interpretable Machine Learning

• In Event: Innovative Applications of Computational Methods

Abstract

Bullying is often seen as a normative aspect of school life, yet it has lasting adverse effects on perpetrators and victims. To better understand its predictors, this study employs data-driven approaches using four waves of the Health Behavior of School Children (HBSC) survey, spanning about 40 countries from 2001 to 2014. Applying interpretable machine learning methods, including random forest algorithms and feature importance rankings, the research identifies key risk factors for bullying perpetration and victimization. The findings highlight that antisocial lifestyle factors strongly predict perpetration, whereas physical and psychological traits are more relevant to victimization. Additionally, these predictors’ relative importance remains stable across time and cultural contexts. By uncovering these consistent patterns, the study enhances the theoretical understanding of bullying behaviors and informs the continued development of targeted, evidence-based policies and interventions to address and prevent bullying in school settings.

Authors

• Heejin Lee, Sam Houston State University

• Pamela Wilcox, Pennsylvania State University

• Won Chang, Seoul National University