CyberGuard: Cyberbullying Detection System using BiLSTM for Enhancing Digital Safety and Justice in Online Communities

The rise in cyberbullying incidents on digital platforms has intensified the need for automated systems that can detect and assess harmful content effectively. This research introduces CyberGuard, a cyberbullying detection system that employs a deep learning (DL) technique based on Bidirectional Long Short-Term Memory (BiLSTM) to identify cyberbullying in social media postings and online comments. The system preprocesses text data using tokenization and padding, putting it into a structured format appropriate for machine learning, utilizing a labeled dataset of online interactions. The BiLSTM model accurately recognizes subtle bullying language patterns by capturing contextual linkages in both forward and backward directions. The model architecture includes an embedding layer, a bidirectional LSTM layer for contextual learning and dense layers for classification. Trained for binary classification, it distinguishes bullying from non-bullying content, with an additional severity classification module to prioritize intervention for high-risk cases. The results demonstrate that the system achieves 87.77% accuracy, precision 88.01%, recall 87.77% and F1 Score 87.81% showcasing its potential as a scalable AI-driven solution to combat digital violence, enhance online safety and support justice mechanisms. By detecting and classifying cyberbullying incidents, CyberGuard contributes to Sustainable Development Goal 16 which promotes peace, justice and strong institutions, by fostering safer digital communities, mitigating online harassment and supporting efforts to build a more inclusive and secure internet space.