An Ensemble Learning Approach for Credit Scoring Problem: A Case Study of Taiwan Default Credit Card Dataset

Abstract

Credit scoring is very important for financial institutions. With the advent of machine learning, credit scoring problems can be considered as classification problems. In recent years, credit scoring problems have been attracted to researchers. They explored machine learning and data preprocessing methods for specific datasets. The difficulties of the credit scoring problem reside in the imbalance of datasets and the categorical features. In this paper, we consider a Taiwan credit dataset which is shared publicly. The small number of studies on this dataset motivates us to carry out the investigation. We first proposed methods to transform and balance the dataset and then explore the performance of classical classification models. Finally, we use ensemble learning, namely Voting which combines the results of some classifiers to improve the performance. The experimental results show that our approach is better than the recent publishes and the Voting approach is very promising.