CBDS-ConvNet: A Cyber-Bullying Detection Model using Convolutional Neural Network

Abstract

In recent years, the increasing reliance on the internet and the integration of social media into daily life have led to significant advancements in various aspects of human activities.However, these developments have also facilitated unethical behaviors, with cyberbullying emerging as a critical concern.Traditional machine learning models for cyberbullying detection face challenges such as vulnerabilities to adversarial attacks and difficulty capturing nuanced or complex contextual information, often resulting in misclassifications.To address these limitations, this research introduces CBDS-ConvNet, a Convolutional Neural Network-based model designed for real time cyberbullying detection and prevention.The model is structured into five key layers: Data Collection, Data Preprocessing, Training, Cyberbullying Detection, and Performance Evaluation.Data from platforms such as Mendeley, Kaggle, and GitHub were utilized, with preprocessing ensuring the text data was clean and suitable for training.The model achieved an accuracy of 77.65%, precision of 56.26%, recall of 63.86%, and an F1 score of 60.20%, outperforming some other machine learning approaches.To further evaluate the robustness of the developed model, it was tested on a synthesized dataset, achieving an accuracy of 91%, precision of 89%, recall of 81%, and an F1 score of 85%.This research shows the capacity of CNNs in tackling the dynamic and complex nature of social media interactions.By enabling real-time cyberbullying detection, the CBD-ConvNet system provides a robust framework for safer online environments, thereby advancing research efforts in the field of cyberbullying prevention.