ENSEMBLE ADABOOST in CLASSIFICATION and REGRESSION TREES to OVERCOME CLASS IMBALANCE in CREDIT STATUS of BANK CUSTOMERS

The Classification and Regression Trees (CART) is a popular classification method. Generally, at a bank, debtors who have delinquent loans (Non-performed Loan/NPL) have a small proportion compared to debtors who have smooth loan (Performed Loan/PL). Standard classification methods CART is not suitable for handling such cases as it is sensitive to classes with a high degree. Hence, additional methods are needed in order to improve classification accuracy in the case of class imbalance. This study aims at determining the results of the classification using the CART and Adaptive Boosting (Adaboost) CART methods on bank loan or credit collectability data where there is class imbalance. The data used for analysis are secondary data in the form of bank debtor credit collectability data with 9 predictor variables and one response variable. Simulations are also conducted to find out the consistency of the results of analysis and general performance of Adaboost CART. The results of this study indicate the accuracy of the classification on the Adaboost CART method can be increased compared to the CART method. This implies that Adaboost can add weights to classifiers which have small misclassifications and can reduce weights on the correctly classified objects. This research can be taken into consideration in choosing the right classification analysis in the case of data with class imbalance. Simulation results confirm that the classification accuracy of Adaboost CART is relatively large, 84.1%.