Financial and Economic Analysis for Assessment of Credit Risk with Explainable Artificial Intelligence

The demand for precise and open credit risk assessment has increased due to the increased reliance on credit-based financial systems. Even though they are interpretable, traditional statistical models frequently fall short in understanding the complex and often unpredictable connections in today's financial data, it's clear that advanced better predictions can be provided by artificial intelligence (AI) algorithms. But these models are not always transparent, and are therefore not useful in a well-regulated financial environment. This study employs explainable artificial intelligence (XAI) implementation is suggested, which combines such methods as SHAP and LIME with gradient boosting and neural networks. To test this, a real-life credit risk dataset is applied to test the ability of the model to estimate loan defaults and measure the creditworthiness of borrowers. The findings indicate that deep learning and ensemble methods are compared to the conventional techniques in accuracy, precision and recall. Simultaneously, SHAP and LIME make the process more transparent by revealing which factors, like debt-to-revenue proportion, income level, credit history, etc., influence it, including their valuesmost influence the model's decisions. The stability of sophisticated models is further supported by robustness tests conducted in noisy environments. By striking a compromise amongst interpretability and accuracy, the suggested method satisfies regulatory requirements and builds stakeholder trust while empowering financial institutions to make equitable, open, and trustworthy credit risk judgements.

Перевод пока недоступен