Comparison of Machine Learning and Deep Learning Models for Detecting Cyberbullying

• Published in: 2024 International Visualization, Informatics and Technology Conference (IVIT) pp. 138 - 144
• Main Authors: Lo, Kevin Alexander, Briant Joe, Cornelius, Philip, Samuel, Hidayaturrahman
• Format: Conference Proceeding
• Language: English
• Published: IEEE 07.08.2024
• Subjects: Accuracy; AI-based Detection Systems; Biological system modeling; Computational modeling; Content Moderation; Convolutional Neural Network (CNN); Convolutional neural networks; Cyberbullying; Data models; Deep learning; Machine Learning; Natural Language Processing (NLP); Online Communication; Online Safety; Random Forest; Random forests; Real-time Detection; Real-time systems; Recurrent Neural Network (RNN); Sentiment Analysis; social media; Support Vector Machine (SVM); Support vector machines
• DOI: 10.1109/IVIT62102.2024.10692892

Since social media has grown, cyberbullying has become a widespread problem that seriously damages both people and society. This work looks at how deep learning (DL) and machine learning (ML) models might be used to identify cyberbullying on social media. We use advanced DL architectures like Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN) together with ML methods like Random Forest and Support Vector Machine (SVM) to analyze textual data and find cases of cyberbullying. A dataset of 39,999 tweets classified as cyberbullying is used in the study; to guarantee data quality, it is thoroughly preprocessed, including text normalization, tokenization, and vectorization. We show that, in terms of accuracy, precision, recall, and F1-score, DL models-especially CNN and RNN-far outperform conventional ML models. Higher computational complexity and training time notwithstanding, DL models show better capacity to identify complicated and context-rich cyberbullying cases. The results indicate that the RNN model achieved the highest accuracy of 84.71%, followed closely by the CNN model with 84.01%. The discussion highlights DL models' superior ability to capture complex patterns in textual data, making them more effective for cyberbullying detection, although they require significant computational resources. This study emphasizes the potential of incorporating AI-driven solutions for real-time monitoring and mitigation of cyberbullying, advocating for further optimization of these models and their integration into practical applications to foster safer online environments. Future development will focus on efficiency optimization and the deployment of these models for real-time cyberbullying detection.