Predicting cyberbullying victimisation in emerging markets and developing countries using the Global School-Based Health Survey

Abstract

Objectives

This study aimed to identify predictors of cyberbullying victimisation among adolescents and develop predictive models to support early intervention strategies.

Methods

Data from the Global School-based Health Surveys (2017–2021) were analysed, focusing on emerging markets and developing countries. A simple random sampling strategy was used to ensure equal representation across countries. A multivariable logistic regression model was applied to 26 variables to identify significant predictors of cyberbullying victimisation. Subsequently, machine learning techniques were used to develop predictive models.

Results

This logistic regression model was statistically significant (χ2(26)=507.96, p < 0.001), explaining 19.3 % of the variance with an AUROC of 0.758 (95 % CI, 0.739 to 0.778). Twelve variables, including being bullied on school property, female gender, peer victimisation, early sexual debut, alcohol consumption, and suicidal ideation, were identified as significant predictors. The best-performing predictive model, a randomly over-sampled random forest classifier, achieved 82 % accuracy and an AUROC of 0.83 (95 % CI, 0.81 to 0.85).

Conclusions

The study highlights key predictors of cyberbullying victimisation and demonstrates the potential of machine learning in developing accurate predictive models. However, reliance on self-reported data may introduce biases. Future research could integrate diverse data sources to enhance model accuracy and reliability.