Over the past decade, digital communication has reached a massive scale globally. Unfortunately, cyberbullying has become prevalent, with perpetrators hiding behind the mask of relative internet anonymity. In this work, efforts were made to review prominent classification algorithms and also to propose an ensemble model for identifying cases of cyberbullying, using Twitter datasets. The algorithms used for evaluation are Naive Bayes, K-Nearest Neighbors, Logistic Regression, Decision Tree, Random Forest, Linear Support Vector Classifier, Adaptive Boosting, Stochastic Gradient Descent and Bagging classifiers. Through experimentations, comparisons were made with the classifiers against four metrics: accuracy, precision, recall and F1 score. The results reveal the performances of all the algorithms used with their corresponding metrics. The ensemble model generated better results while Linear Support Vector Classifier (SVC) was the least effective of all. Random Forest classifier has shown to be the best performing classifier with medians of 0.77, 0.73 and 0.94 across the datasets. The ensemble model has shown to improve the results of its constituent classifiers with medians of 0.77, 0.66 and 0.94, as against the 0.59, 0.42 and 0.86 of Linear Support Vector Classifier.
Consensus-Based Ensemble Model for Arabic Cyberbullying Detection
