According to the World Health Organization (WHO), malignant tumor (cancer) is one of the leading causes of mortality worldwide. Among all types of malignant tumors, malignant breast tumor (breast cancer) is the most common malignant tumors found, especially in women. One of four ways to distinguish benign breast tumor from malignant ones is by doing Fine Needle Aspiration (FNA) test. This method is simple, quick, inexpensive, and can be performed either in outpatients or inpatients. Although FNA is much more preferred than the other methods, the accuracy of FNA varies widely. Therefore, this research is conducted to find the best model to classify malignant breast tumors and benign breast tumors, based on data from FNA test by evaluating value of Area Under the Curve (AUC) of three algorithms, Extreme Gradient Boosting (XGBoost), Support Vector Machine with Radial Basic Function kernel (SVM-RBF), and Multi-layer Perceptron (MLP). The breast cancer data set used in this research is obtained from Wisconsin Breast Cancer (Original) Data set available in the UCI Machine Learning Repository. The experiment shows that SVM with eliminating missing value data set achieved the best result, with AUC value 99.23 and cost $2,740.20.