Cyberbullying Detection Using Linear Discriminant Analysis and Glove Feature

Abstract

Because of the exponential rise in social media users, cyberbullying has developed as a form of bullying via electronic messages. Given the effects that cyberbullying has on its victims, it is critical to determine the best ways to recognize and stop it. The study used glove features and linear discriminant analysis to improve cyberbullying detection. The Twitter dataset was developed to help in algorithm development and evaluation. The high-dimensional data was projected into a linearly separable feature space that is ideal for downstream classifiers by using Linear Discriminant Analysis (LDA) on the GloVe vectors. The Python environment is used to evaluate research experiments. A number of performance metrics were used, such as the F1 score, accuracy, recall, and precision. The findings show that the Support Vector Machine outperformed the other classifier methods, including Random Forest, SVM, Naïve Bayes, and K-NearestNeigbor, with an accuracy of 99.7%. Using the Twitter dataset, the study found that Glove Feature and Linear Discriminant Analysis (LDA) perform better in extracting bullying tweets.