An Investigation on Vietnamese Credit Scoring Based on Big Data Platform and Ensemble Learning

Abstract

The credit score is a vital indicator that can affect many aspects of people’s lives. However, evaluating credit scores is done manually, so it costs a large amount of money and time. This paper learns from disadvantages of previous research and brings some insights and empirical experiments so as to the advantages of distributed solutions for the problem of credit score in the future. The research compares some feature engineering techniques using a big data platform and ensemble learning methods to find the best solution for predicting the credit score. Since data related to customers’ financial activities grows enormously, a big data platform is necessary to handle this amount of data. In this paper, Spark which is a distributed, data processing framework, is used to save and process data. Some experiments are carried out to compare the effectiveness of feature engineering in this problem. Moreover, a comparative study about the performance of ensemble learning models is also given in this paper. A real-world Vietnamese credit scoring data set is used to develop and evaluate models. Four metrics are used to evaluate the performance of credit scoring models, namely F1-score, recall, precision, and accuracy. The results are promising with the highest accuracy of 72.9% in the combination Gradient-boosted Tree and cleaned data set with removing categorical features. This paper is a foundation for using big data platforms to handle financial data and much future research can be carried out to optimize the performance of this paper.