Examining Machine Learning Approaches for Detecting Cyberbullying in Social Media Content
Journal of System and Management Sciences · 2024
This study investigates machine learning approaches for detecting cyberbullying across textual social media data.Three models -Extra Trees, Random Forest, and XGBoost are evaluated on a labeled dataset of 20k tweets.Results indicate Extra Trees achieves highest accuracy (90%) and AUC-ROC (95%) for classifying cyberbullying vs non-cyberbullying posts.Additionally, lexical analysis of 2000 YouTube comments expands existing knowledge of terms and phrasing markers of online harassment.The research contributes both methodological and practical advances in applying ML to combat rising social media hostility.We make the code and the generated data freely available at https://github.com/jamalalqundus/code-paper-cyberbullying-detection.git for further research.