Possibility of improving credit scoring systems: Using machine learning models

Abstract

This study investigates the potential for improving credit scoring systems in non-bank financial institutions (NBFIs) by applying advanced machine learning techniques. The primary objective is to develop predictive models that more accurately assess credit risk compared to traditional statistical methods. Using real-world, anonymized loan data from a Mongolian NBFI, the study applies and compares three ensemble learning algorithms: Random Forest, XGBoost, and LightGBM. Key variables such as loan amount, repayment history, age, interest rate, and loan term were used to train and validate the models. Among the tested models, XGBoost demonstrated the highest predictive accuracy, with 93% classification performance and strong model robustness as indicated by ROC-AUC and error metrics. The findings highlight that machine learning models, particularly XGBoost, outperform conventional approaches in terms of both accuracy and practical applicability. Moreover, the integration of explainable AI techniques enhances the transparency of the credit scoring process. This research contributes to the localization of modern credit scoring tools in emerging markets and provides a scalable solution to improve financial inclusion and risk-based decision-making in NBFIs