Transparent Loan Approval Prediction: A Machine Learning Approach with SHAP-based Interpretability

Accurate loan approval prediction has become crucial in promoting equitable lending practices and ensuring responsible financial decision-making. Machine learning (ML) models are widely utilized in classifying loan approvals due to their strong predictive accuracy. However, the predictions made by these models are often difficult to explain and trust. This study investigates the CatBoost model using interpretable machine learning methods, specifically focusing on Shapley Additive Explanations (SHAP), to enhance the model’s interpretability, transparency, and understanding of factors influencing loan approval decisions. We implemented the Catboost Model and compared its predictive performance with eight ML models, including XGBoost, LightGBM, Gradient Boosting, Random Forest, Extra Trees, Decision Tree, AdaBoost, and BaggingClassifier for predicting loan approval status. The Catboost model demonstrates impressive accuracy, achieving a remarkable accuracy of 99.85%, outperforming other classifiers such as XGBoost at 99.38% and Bagging-Classifier at 99.25%. The outputs of the most accurate model, CatBoost, were explained using model-agnostic explanation techniques called Shapley additive explanations (SHAP). The SHAP method revealed that Risk Score, Debt-to-Income Ratio, and Total Debt-to-Income Ratio are the three key features influencing loan approval. Additionally, it demonstrated how each feature impacts the predictions made by the CatBoost model. Our primary finding highlights the significance of SHAP in enhancing the interpretability and explainability of loan approval models, providing financial institutions with a valuable tool to ensure more transparent, reliable, and informed decision-making in the lending process.