Analyzing the Performance of Machine Learning and Deep Learning Models in Detecting Cyberbullying Comments

With the proliferation of social media platforms, there has been a concerning escalation in cyberbullying and abusive language. Detecting such toxic comments in large volumes of user-generated data presents challenges. This research conducts a comparative analysis to evaluate the impact of natural language processing techniques on identifying cyberbullying. Using a dataset of 1,59,000 comments labelled across seven classes of toxicity, conventional machine learning and deep learning models are benchmarked across performance metrics. The study aims to quantify algorithmic capabilities to accurately classify bullying comments. The study finds that Support Vectors with a linear kernel surpass Logistic Regression in accurately identifying toxic comments. For deep learning techniques, the transformer models (BERT and Distil-BERT) deliver the highest performance among neural network architectures tested. The empirical evaluation provides insights into leveraging computational linguistics for automating the detection of online bullying at scale.