Abstract

Cyberbullying has emerged as a pressing issue in the digital era, particularly within Arabic-speaking communities, where research remains limited. This study investigates the detection of Arabic cyberbullying on social media using both traditional machine learning (ML) and deep learning (DL) techniques. A publicly available dataset of Arabic tweets was used to train and evaluate several ML models (SVM, NB, LR, and XGBoost), alongside a recurrent neural network (RNN). The results demonstrate that the RNN significantly outperforms classical ML models, highlighting the efficacy of DL in accurately identifying abusive content in Arabic text. These results emphasize the necessity of incorporating linguistically rich data and advanced neural architectures to improve cyberbullying detection systems in low-resource languages such as Arabic.

Key words: Machine Learning Algorithms, Arabic Tweets, Deep Learning Techniques, Recurrent Neural Network, Cyberbullying.