Analysis of the Performance Comparison between Random Forest and SVM RBF in Detecting Cyberbullying on Imbalanced Data with the SMOTE Approach

Abstract

Cyberbullying has emerged as a growing threat with the widespread adoption of social media, creating significant risks to online safety. Automatic detection of such behavior remains challenging, particularly when the training dataset is highly imbalanced. This study presents a comparative analysis of Random Forest and Support Vector Machine with Radial Basis Function kernel (SVM RBF) for cyberbullying detection, incorporating the Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance. The experiments utilized a publicly available, manually annotated dataset containing 47,693 English-language tweets from global users, labeled as cyberbullying or non-cyberbullying. Performance was evaluated using accuracy, precision, recall, and F1-score. Results indicate that Random Forest achieved the highest performance before SMOTE (accuracy = 88.52%, precision = 89.07%, recall = 94.00%, F1-score = 91.49%), while SMOTE improved recall for both algorithms but reduced accuracy and precision. These findings highlight that the choice of algorithm and effective handling of class imbalance are critical for enhancing the reliability of automated cyberbullying detection systems, thereby enabling more effective content moderation and safer online environments.