SUMMARY
In social media, it is found that hate speech is conveyed in the form of text, images and videos, as a result it can provoke certain people to do things that are against the law and harm other person. Therefore, it is necessary to make early detection of hate speech by utilizing machine learning algorithms. This study is to analyze the level of accuracy, precision, recall and F1-Score of 3 kinds of algorithms (SVM, XGBoost, and Neural Network) in the classification of hate speech, using datasets sourced from public hate speech on Twitter in Indonesian. The results of the analysis show that the SVM algorithm has a level of accuracy (83.2%), precision (83%), recall (83%) and F1-score (83%), SVM occupies the highest level compared to XGBoost and Neural Network, so the SVM algorithm can be considered for use in hate speech classification