Integrating AI in savings and credit management system to enhance financial loan risk assessment.

Abstract

The financial sector is experiencing a significant transformation through the integration of Artificial Intelligence (AI) technologies, particularly in credit risk assessment and financial decision-making processes. Savings and Credit Cooperatives (SACCOs), such as KIZIGURO Savings and Credit Cooperative (KISACCO) in Rwanda, play a crucial role in promoting financial inclusion by providing credit facilities to underserved populations. However, traditional loan risk assessment methods within SACCOs predominantly rely on basic salary and asset-based guarantees. These static approaches often fail to accurately assess members’ true repayment capacity, leading to increased loan defaults, exclusion of eligible borrowers, and limited financial inclusion. This research made the integration of AI and machine learning (ML) models into KISACCO’s loan management system to revolutionize credit risk assessment. By analyzing members' historical and current transactional behaviors, the AI-driven model predicts eligible loan amounts and probability levels. This approach moves beyond the limitations of traditional methods, offering a more precise, datadriven, and personalized credit evaluation framework. The study aims to develop an AI-based predictive model that enhances loan approval accuracy, minimizes default risks, and improves operational efficiency using Random Forest Model and dataset from KISACCO. Based on the comparative evaluation of the four machine learning algorithms: Logistic Regression, Decision Tree, Random Forest, and Support Vector Machine (SVM), using KISACCO member transaction records, the Random Forest model demonstrated the highest overall performance with an accuracy of 0.983, precision of 0.984,recall of 0.988,F1-Score of 0.986 and AUC-ROC of 0.94 . The Decision Tree model followed, performing moderately well with an accuracy of approximately 0.88, precision of 0.86, and recall of 0.84, reflecting its ability to handle complex patterns but also its susceptibility to overfitting. Support Vector Machine (SVM), known for its simplicity and interpretability, achieved an accuracy of around 0.83, with precision of 0.82, and recall of 0.78, indicating fair performance in identifying credit risk but limited in capturing nonlinear relationships. Logistic Regression, while powerful in theory, performed slightly lower with an estimated accuracy of 0.81, precision of 0.79, and recall of 0.76, possibly due to its sensitivity to parameter tuning and feature scaling. Overall, the Random Forest model proved to be the most reliable and effective tool for predicting loan default risk within the KISACCO dataset. Furthermore, the model is expected to increase customer satisfaction by providing fairer loan evaluations and promoting responsible lending practices. Ultimately, this research will contribute to the development of an intelligent decision-support system for SACCOs, facilitating smarter, more inclusive, and efficient financial loan management.