Social media is growing rapidly at the moment and provide convenience to communicate. But such convenience widely misused to treat other people with not decent before the entire internet community commonly called cyberbullying. If cyberbullying fail to prevent, it will be difficult to track down and deal with it. One of the main weapons to prevent acts of cyberbullying is to perform detection on social media. Detection of cyberbullying can be done by determining whether a post offend the sensitive topic of a personal nature such as racist or not. By determining the related words such sensitive topics and filter sentiment, cyberbullying tweet detection is done by using the method of classification Hyperpipes, Tree-based J48, and SVM. The results show that the algorithm hyperpipes and decision tree produces the best evaluation results with the accuracy of 85.32% and 86.24%.