Explainable Artificial Intelligence Credit Risk Assessment using Machine Learning

Transparency in lending decisions is achievable: The paper demonstrates that AI systems can make credit decisions while remaining interpretable to stakeholders. Using techniques like SHAP and LIME, the system generates visual explanations for individual loan decisions, meaning applicants and regulators can understand why a loan was approved or denied rather than facing a "black box" verdict.

Multiple models improve fairness outcomes: By comparing three different algorithms (XGBoost, LightGBM, and Random Forest), the researchers identified that LightGBoost offered the best "trade off between approval and default rates." This suggests that model selection itself is a responsible AI decision—choosing the algorithm that balances business needs with equitable lending practices matters.

Technical safeguards address real-world lending problems: The paper addresses class imbalance (where defaults are rarer than approvals) using SMOTE, which helps prevent the AI from being biased toward approving loans simply because defaults are uncommon in training data. This technical choice has direct implications for fair lending practices.

Explainability enables accountability: By generating "applicant-specific XAI visual reports," the system creates an audit trail for lending decisions. This supports regulatory compliance and allows institutions to demonstrate they're making decisions based on relevant factors rather than protected characteristics like race or gender.