Identification and Detection of Cyberbullying on Facebook Using Machine Learning Algorithms

The use of social media platforms such as WhatsApp, Facebook, Instagram, and Twitter has facilitated efficient, effective, and frequent communication amongst people. Despite the numerous benefits associated with these social media platforms, they have also resulted in cyberbullying, which frequently occurs while using these networks. Cyberbullying is known to be the cause of some serious health, emotional, psychological, and social issues among social media users. With damages done globally with this social media threat, creating a way to identify and detect it is very significantly important. Against this backdrop, this paper takes a look at unique features obtained from the Facebook dataset and utilized machine learning algorithms to identify and detect cyberbullying posts and subsequently notify the internet users of some undesirable features they should desist from when they are being harassed or bullied in cyberspace. The algorithms used are naive Bayes and k-nearest neighbor. The study also uses a feature selection algorithm, namely the x(2) test (chi-square test) to select important features leading to improvement in the classification performance. The result of the study indicates the detection of cyberbullying on Facebook with a high degree of accuracy with the selected machine learning algorithms along with the chosen metrics for performance evaluation. Specifically, the k-nearest neighbor performed better when compared to naive Bayes classifier with much improvement in the performance and classification time.