Bank Loan Classification of Imbalanced Dataset Using Machine Learning Approach

Abstract

Before giving loans to borrowers, banks decide whether the borrower is bad (defaulter) or good (non-defaulter). The prediction of borrower status whether the borrower will be a defaulter or a non-defaulter is not an easy task to the loan providing entity. In machine learning, building an automated loan default classification system is an optimization problem with an ultimate objective of improving loaner classification in loan decision making. However, this problem becomes difficult when there is a profile of imbalanced data since the classifier may misclassify the rare samples from the minority class. To solve this problem, we used a Modified Synthetic Minority Oversampling Technique (MSMOTE). It is an oversampling technique where synthetic data of the minority class is generated to balance with the majority class. This is combined with ensemble classifier technique to further improve the performance of bank loan prediction systems. MSMOTE is a variant of Synthetic Minority Oversampling Technique (SMOTE) algorithm. Bagging- and boosting- based ensemble techniques are applied on the imbalanced dataset to improve the performance of loan prediction. The dataset is gathered from Kaggle to validate the proposed scheme. Experimental results show that, among others, the proposed model, MSMOTE, when combined with adaptive boosting resulted in 95% of precision and accuracy. Whereas, MSMOTE combined with Bagging and Random Forest resulted in 99% of precision and accuracy.