KMAN Publication Inc. (KMANPUB) Journal of Adolescent and Youth Psychological Studies (JAYPS) 2981-2526 7 Serial Number 42 2026 02 10 Predicting Cyberbullying Perpetration via Random Forest Modeling of Moral Disengagement and Empathy Deficits Predicting Cyberbullying Perpetration via Random Forest Modeling of Moral Disengagement and Empathy Deficits 1 11 10.61838/kman.jayps.5086 EN Journal Article 2025 09 25 <table> <tbody> <tr> <td> <p><strong>Objective:</strong> The present study aimed to predict cyberbullying perpetration among adolescents using Random Forest modeling of moral disengagement mechanisms and empathy deficits.</p> <p><strong>Methods and Materials: </strong>This cross-sectional quantitative study was conducted among 742 secondary school students (ages 13–18 years) from three provinces in South Africa using multi-stage cluster sampling. Participants completed validated self-report instruments measuring cyberbullying perpetration, moral disengagement, and empathy deficits, alongside demographic indicators and daily internet usage. Data were screened, cleaned, and randomly divided into training (70%) and testing (30%) datasets. A Random Forest classifier with 500 trees was trained to distinguish high versus low cyberbullying perpetration. Hyperparameters were optimized using cross-validation. Model performance was evaluated through accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC-ROC). A logistic regression model was estimated as a baseline comparator. Variable importance indices and partial dependence plots were generated to examine predictor contributions and non-linear interaction patterns.</p> <p><strong>Findings: </strong>Inferential analyses indicated that moral disengagement and empathy deficits were significant predictors of cyberbullying perpetration (p &lt; .001). Male students reported higher levels of cyberbullying and moral disengagement (p &lt; .001). The Random Forest model outperformed logistic regression, achieving superior classification accuracy (0.86 vs. 0.74) and AUC-ROC (0.91 vs. 0.78). Variable importance metrics identified overall moral disengagement, dehumanization, and attribution of blame as the strongest predictors, followed by empathy deficits. Partial dependence analysis revealed non-linear threshold effects, with sharp increases in predicted cyberbullying probability at higher levels of moral disengagement, particularly when combined with elevated empathy deficits.</p> <p><strong>Conclusion: </strong>The findings demonstrate that cyberbullying perpetration is most strongly predicted by moral disengagement mechanisms and empathy deficits.</p> </td> </tr> </tbody> </table> https://www.journals.kmanpub.com/index.php/jayps/article/download/5086/9177