Ensemble Approach with Hyperparameter Tuning for Credit Worthiness Prediction

Data mining algorithms has a wide application in banking domain. Classification algorithms are the one of the popularly used algorithms in the banking sector. One of the major applications of classification algorithms in banking sector is predicting the credit worthiness of the borrower. In this research the dataset is taken from prowessIQ and the dataset is highly imbalanced. The researchers have used data sampling techniques to balance the data. Our approach is to use the ensemble classifiers which are performing pretty well even on unbalanced datasets. Three ensemble techniques are studied in this paper namely Random Forest (RF), XGB and LGBM classifiers. Hyper parameter tuning is done to improve the accuracy of the model. With best possible hyperparameters RF has given an accuracy of 90.2%, XGB has given an accuracy of 92.7% and LGBM has given an accuracy of 92.9%. LGBM has better classified borrowers as worthy and non-worthy even when the dataset is unbalanced.