Accounts are a collection of numbers commonly used for all transactions in the banking world, from saving, withdrawing cash, to checking account balances either directly or online using m-banking. In bank accounts and the process of opening them, there are various kinds of criminal acts committed by individuals and groups. As a bank's obligation to prevent crime so that it can provide trust to the community. In an effort to prevent criminal acts of fraud can be solved using data mining techniques, namely classification. The purpose of classification is to predict the class label of an object based on existing attributes. The classification methods used in this research are extreme gradiant boosting (XGBoost) and random forest on 22029 records. In the classification process, this study uses a percentage ratio of 90% train data and 10% test data and tuning parameters processed by randomized search cross validation. The research stages start from preprocessing to evaluation and get a train score of 99.50% and a test score of 99.59% for extreme gradiant boosting (xgboost) while random forest gets a train score of 99.46% and a test score of 99.59%. These results show that the classification results of extreme gradiant boosting (XGBoost) are better than random forest.

No abstract