Development of Machine Learning-Based Customer Credit Scoring Models: Analysis of Bank Melli Iran Data for Credit Risk Prediction

Abstract

The aim of this study was to develop and evaluate customer credit risk prediction models using machine learning algorithms based on real-world data from Bank Melli Iran. In this research, a dataset consisting of 24,860 individual customers with 42 financial, demographic, behavioral, and credit-related variables was utilized. The default rate of the dataset was 14.8%, providing an appropriate basis for evaluating classification models. After conducting preprocessing procedures, including outlier removal, missing value imputation, and data normalization, five models—Logistic Regression, Decision Tree, Random Forest, XGBoost, and Multilayer Perceptron Neural Network—were trained. Model performance was evaluated using indicators such as Accuracy, Precision, Recall, F1-Score, and Area Under the Curve (AUC). The results demonstrated that the XGBoost algorithm achieved the best performance in predicting customer default probability, with an accuracy of 86.1% and an AUC value of 0.912. Variable importance analysis also revealed that the debt-to-income ratio, number of overdue installments, history of payment delay, and average account balance were among the most influential factors in determining credit risk. The findings indicate that machine learning models, particularly gradient boosting-based methods, can significantly enhance the accuracy of banking credit scoring systems.