Cyberbullying Detection: An Advanced Machine Learning and GANs Approach
Batoul Rida Haidar, Mariam Ezzeddine · 2024
The increase of cyberbullying in the digital era creates enormous issues, demanding novel techniques for identification and prevention. This study explores a cyberbullying dataset from Kaggle including various types of online hostility for the goal of cyberbullying identification using advanced methods of machine learning such as Support Vector Machine (SVM) and fine-tuned BERT. The work handles an unbalanced dataset with thorough NLP preprocessing steps enhancing the model’s accuracy to 90% for SVM and 97% for fine-tuned BERT. To achieve this enhancement, a generative adversarial network (GAN) is used in the experiment to produce 10,000 additional bullying texts, diversifying the dataset and increasing model performance. These developments highlight the importance of machine learning in cyberbullying detection and open the way for future research in Arabic language cyberbullying detection.