Sentiment Analysis of Cyberbullying Using Machine Learning

Social media has become an extremely popular communication tool, especially in Indonesia, with the number of active users reaching 167 million in 2024. However, the popularity of social media also brings risks, such as cyberbullying, which has emerged as an issue with serious psychological, social, and even physical consequences for victims. Therefore, effective detection of cyberbullying is needed, one approach being the use of machine learning. While numerous studies have been conducted, further research is required to directly compare these machine learning methods to identify the strengths and weaknesses of each. This study aims to compare machine learning methods, such as SVM, KNN, Naive Bayes, and Logistic Regression, for detecting cyberbullying. Based on the experimental results, Naive Bayes demonstrated the best performance with an average accuracy of 91%, followed by Logistic Regression with an average accuracy of 89%. A confusion matrix and K-Fold cross-validation were also used to evaluate model performance, with Naive Bayes and Logistic Regression showing the highest consistency. Thus, it can be concluded that the probabilistic approach of Naive Bayes is more suitable for detecting cyberbullying in the dataset used.