Detecting Cyberbullying Using Machine Learning Approaches

This paper investigates the identification of cyberbullying using machine learning approaches, especially using Natural Language Processing (NLP) techniques. Particularly for teenagers, cyberbullying—defined as negative behaviors carried out via digital media—offers serious threats. Starting with an analysis of the frequency and consequences of cyberbullying—noteably $37 \%$ of teenagers have come across this behavior online—the study moves on. The paper uses XGBoost, Adaptive Boosting (AdaBoost), and Random Forest among several machine learning techniques to classify text data taken from social media. More than 432,000 entries in the dataset fall under both non-cyberbullying and cyberbullying. Text preprocessing, TF-IDF feature extraction, and accuracy, precision, recall, and F1-score evaluation metrics comprise the used approaches. With 5000 features and a $95 \%$ accuracy rate, Random Forest shows that it outperforms other methods and so highlights the effectiveness of machine learning models in tackling the cyberbullying problem. Applying these techniques to well-known social media platforms shows that they could significantly increase real-time detection capacities, so offering a possible way to reduce the prevalence of online harassment.