Cyberbullying Detection on Twitter: A Comparison of One-Step vs. Two-Step Classification Approaches
Davin Edbert Santoso Halim, Felicia Andrea Tandoko, Lili Ayu Wulandhari, Ghinaa Zain Nabiilah · 2025
Cyberbullying detection on social media platforms, especially on Twitter, is crucial for ensuring online safety and addressing harmful online behaviors. As cyberbullying becomes more common, automatic detection systems are essential for identifying abusive content and protecting users. This study compares two classification approaches: a one-step approach and a two-step approach, for detecting and categorizing cyberbullying content. The one-step approach classifies tweets into various categories of cyberbullying, while the two-step approach first determines whether a tweet contains cyberbullying or not, and then classifies it into specific categories if it does. We evaluate the performance of both models using three machine learning algorithms: Random Forest, XGBoost, and SVM, all optimized through hyperparameter tuning. The results show that the two-step approach outperforms the one-step approach, achieving an overall accuracy of $91 \%$ compared to $84 \%$ for the one-step approach. XGBoost performs the best for binary classification, while SVM performs best in multiclass categorization. These results highlight the potential of the two-step approach in providing a more detailed understanding of cyberbullying and its types.