Credit Risk Prediction using Extra Trees Ensemble Method

Abstract

Credit risk is the measurement of a person’s likelihood of being able to pay back a loan borrowed from the bank in the future. If the bank borrower can pay the money back later only then will the bank lend money to the borrower otherwise not. Because the entire computation of the borrower’s asset is done manually without the assistance of cutting-edge tools or technologies, the work of determining whether to lend money to the borrower or not is laborious, time-consuming and subject to human errors. A bad prediction could cost the bank a lot of money if the borrower is unable to repay the money that was lent to them. To overcome all these problems this paper proposes an expert system named as Expert System for Credit Risk Prediction using SMOTE and ENN (ESCRPSE) which uses the combination of an oversampling technique known as Synthetic Minority Oversampling Technique (SMOTE) and an undersampling technique known as Edited Nearest Neighbor (ENN) to deal with class imbalance problem and uses ensemble bagging technique Extra Trees (ET) to make the prediction. Two datasets have been used for the proposed paper. The accuracy and fl-score achieved by the proposed model ESCRPSE is compared with different single-classifier based models and various ensemble models present in the literature and it is observed that the proposed model outperforms these models greatly.