Detecting Social Media Cyberbullying on Bangla Language Using Machine Learning

Abstract

On the internet, the number of cyberbullying importunity in the Bangla language is adding in a noteworthy way. All kinds of people like men, women and youths are being the victims of cyberbullying substantially through social medias. There's hardly the system of discovery on the cyberbullying in Bangla language. My ideal is to descry cyberbullying and to argue out of the bullying using machine learning. To complete this ideal there's the need of Bangla dataset, but unfortunately this dataset is veritably rare to find. So I collected the data from Youtube, Facebook etc. using some scrapper tools. The dataset is labelled as cyberbullying “ YES” or “ NO”. Machine learning is the stylish way of approach for my work. I've used many a type of algorithms like Natural Language Processing (NLP), Logistic Regression (LR), Multinomial Naïve Bayes (MNB), Support Vector Classifier (SVC), Random Forest Classifier (RFC), Decision Tree Classifier, KNeighbors Classifier, AdaBoost Classifier, Bagging Classifier, ExtraTreeClassifier, GradeintBoosting Classifier, XGB Classifier. After applying all these algorithms, the exactitude is plant in Logistic Regression (LR) 89.81%, Multinomial Naïve Bayes (MNB) 89.38%, Support Vector Classifier (SVC)90.0%, Random Forest Classifier (RFC) 89.91%, Decision Tree Classifier 86.39%, GradeintBoosting Classifier 89.81%. And the maximum exactitude in Support Vector Classifier (SVC), Which is 90.0%