Application of Ensemble Tree Algorithm for Installment Payment Arrears Prediction at Makmur Bersama Credit Union

Abstract

One of the widely used machine learning techniques is the ensemble tree method, which is a combination of several classification trees where the final decision is based on the combined predictions of each tree. This approach produces better accuracy than a single classification tree. Two common methods used in the ensemble tree technique are boosting and bagging. This research will predict the status of installment payments at CU Makmur Bersama Credit Union. The method used is the bagging tree method, namely random forest and boosting, namely AdaBoost. To get optimal results, hyperparameter tuning is also carried out. The results showed that the boosting and bagging ensemble tree methods were able to handle the classification of cooperative loan installment payment status better than the distance approach, namely kNN (single classification).  The performance of the boosting ensemble tree with the AdaBoost model has an accuracy of 72.89% better than the bagging ensemble tree with the random forest model whose accuracy is 72.08%.