A DEEP LEARNING APPROACH FOR CREDIT SCORING

Abstract

Granting credit to customers is the core business of a bank. Hence, banks need adequate models to decide to whom to approve a loan. Over the past few years, the usage of deep learning to select appropriate customers has attracted considerable research attention. However, the data shortage, type of features, and data imbalance could decrease deep learning model performance from the accuracy perspective. This study aims to build a classifier for credit scoring based on deep learning. We use a credit scoring dataset publicly available on the UC Irvine Machine Learning Repository, a source of machine learning datasets commonly used by researchers. The model architecture is designed to be suitable for two kinds of input features, categorical and numerical ones. Our proposed model gave a relatively high accuracy among recent deep-learning-based models on the same dataset. We also consider the bank profit when applying the model, which is the ultimate goal of lenders. We found that if the banks use our model, they could gain a significant profit.