Cyberbullying Detection on Social Media using a Hybrid Machine Learning Model

Cyberbullying has increasingly become a significant issue in digital environments, Causing psychological harm among users. Effectively identifying harmful content is difficult due to the complexity of language and the constantly evolving slang. This research tackles the shortcomings of conventional detection techniques by incorporating capsule networks with BERT-based Transformers to enhance contextual comprehension. The proposed hybrid model captures hierarchical connections and deep semantic features to boost classification accuracy. Experimental results indicate that the model achieved 99.0%of accuracy, 97.83% precision, 80.0% of recall, and f1-score 98.9% surpassing existing methods. Future improvements will be involve expanding the dataset and fine-tuning sentiment analysis methods for excellent adaptability.