Process Innovation for Credit Scoring Using Machine-Learning Approach for Small Financial Institutions

Abstract

Lending is important activity for overall economy, in which it helps fund investment for entrepreneurs to produce goods and services and also helps speed up consumption for the economy. However, credit default, which is the credit to the borrowers who cannot pay back the loan, can create draw back to the economy and cause higher cost of borrowing to all borrowers as financial institutions will increase interest to cover loss from default customers. Then, managing credit default risk is the key success for financial institutions and credit scoring is one of the tools that financial institutions use to manage their credit default risk for consumer loans. Machine Learning with supervised learning technique has been used to develop credit scoring model to classify good customers from default customers for many years. However, due to its complexity and less friendly than other techniques i.e. statistic or judgement method, the use of machine learning to build credit scoring model is limited to only large financial institutions, especially in Thailand market. This study aims to focus on building credit scoring model using supervised learning for medium to small financial institutions in Thailand, in which there are more limitations than large financial institutions in terms of size and quality of credit dataset. This study also focus on imbalanced data problem between majority and minority class of the dataset, which normally number of good customers always dominates number of default customers.