Performance of a Loan Repayment Status Model Using Machine Learning

Due to Islamic financial management, Islamic cooperative plays an important role in providing financial credit to people in the southernmost of Thailand. The current study aimed to investigate the model for predicting loan repayment status of As- Siddeek Islamic Cooperative Limited in Songkhla Province. Three feature selection techniques were used to examine the association between factors and loan repayment status including Information Gain, Gini Index, and Chi-Squared test. Machine Learning Algorithms namely Logistic Regression, Decision Tree, and Random Forest were used to fit the model. The results showed that the similar important factors from those three techniques were cooperative branch, loan objective, member age, and previous loan amount. The accuracy of prediction models for Decision Tree, Logistic Regression, and Random Forest were 63.1%, 60.9%, and 60.0%, respectively.

I. INTRODUCTION

As- Siddeek Islamic Cooperative Limited, Songkhla Province, was established in 1994. Currently, the cooperative has seven branches, providing financial services to members of the cooperative, such as member shares, Wadiah (savings) deposits, Mudharabah deposits (joint investment) , loans, and welfare services. The cooperative has 25,151 members, a total of assets more than 1,661 million Baht, and 4,922 loan contracts per year. It was accounted for 1.72% of the total loan amounts which up to 355 million Baht per year. In 2019, the cooperative recorded bad debts and doubtful debts of 6.2 million Baht [1].