Cyberbullying has emerged as a critical issue with the widspread use of social media and online communication platforms, leading to harmful psychological and social impacts. Existing moderation systems often rely on manual monitoring or isolated detection methods, which are inefficient in handling large volumes of real-time data. To address this challenge, we developed CyberGuard, an AI-powered cyberbullying detection system capable of identifying and analysing harmful textual content in real time. The proposed system classifies user messages into bullying and non-bullying categories using machine learning techniques, specifically TF-IDF for feature extraction and the Naive Bayes classifier for efficient text classification. The model was trained on labelled textual datasets with preprocessing steps such as tokenisation, stop-word removal, and normalisation to improve accuracy and consistency. Additionally, the system provides confidence scores and severity levels to enhance the interpretability of predictions and support decision-making. The platform integrates a Node. js-based backend for handling API requests and model execution, along with a React-based frontend that enables real-time chat monitoring and interactive visualisation of results. This approach improves accessibility and usability by allowing users and administrators to detect and respond to cyberbullying incidents efficiently within a single unified system.
Lakshmi et al. (Wed,) studied this question.