The use of counterfactual explanations has become more crucial for ensuring fairness and interpretability for automated decision-making systems, particularly in high-risk domains such as healthcare and finance. To address this issue, this research work describes a loan approval prediction system employing a K-Nearest Neighbors (KNN) approach coupled with the DiCE (Diverse Counterfactual Explanations) toolkit. The key goal of this research work is to develop realistic and practical explanations for the decision boundaries of interpreted machine learning systems that can be understandable by humans. The proposed approach focuses on interpretability as a key requirement for maintaining predictive efficiency by carefully altering key variables like income, loan amount, and CIBIL score. Counterfactual reasoning has been applied by numerous researchers in multiple domains for different applications; this research work broadens the horizon by demonstrating the successful combination of the DiCE toolkit with actual financial datasets. We use standard classification metrics to measure how well the model works, and we compare the generated counterfactuals to the actual model outcomes to show how clear the system is. This strategy helps users find the smallest changes that need to be made to change a loan decision from rejection to approval, which is in line with the ideas behind explainable artificial intelligence (XAI).
Patel et al. (Wed,) studied this question.