A Comparative Study of Machine Learning Approaches for Cyber bullying Detection in Digital Forums

Social connections developed under narrow cultural bounds, such as physical locations, prior to the invention of information and communication technologies (ICT). Recent advances in communication technology have greatly surpassed the spatial and temporal limitations of traditional communications. But the improper use of social technology, like social media platforms, has given rise to a brand-new kind of aggressiveness, violence, and abuse that only takes place online. This analysis highlights a new type of social media website behavior. We thoroughly evaluate cyberbullying (CB) prediction models. We examine cyberbullying prediction models in great detail and pinpoint the key problems that surround their creation on social media. This paper describes the overall process for identifying cyberbullying in depth and, more importantly, it summarizes the method. The aim of this effort is to build an efficient method for spotting online bullying and abusive messages by combining NLP and Machine Learning (ML) technologies. For this, I have gathered 2 datasets (DATASETS 1 AND 2) from various sources. Dataset 1 has 10,000 comments, whereas Dataset 2 has 20,000 comments. Tokenization, stop words, Bag of Words (BoW), and six machine learning algorithms—Logistic regression (LR), SVM, Random forest (RF), Decision Tree (DT), Naive Bayes (NB), and XG Boost (XG)—have all been used in NLP. I have trained and each of the 6 ML algorithms, and I have analyzed the outcomes for each independently. Results demonstrate that RF is more effective for both datasets, showing 99.39% accuracy, 99.31% precision, 99.93% recall, 99.61% F1-score, and 97.32% specificity for Dataset 1 and 98.86% accuracy, 97.7% precision, 99.48% recall, 98.58% F1-score, and 98.45% specificity for Dataset 2 accordingly.