Machine Learning and Explainable AI for Credit Risk Assessment: Enhancing Accuracy, Reducing Default Errors, and Maintaining Compliance.
0 viewsAbstract
This study is aimed to examine ways to improve traditional credit risk assessment processes by integrating machine learning algorithms. Conventional methods are characterized by linear analysis that are known for misclassifying borrowers and are prone to high default errors. The primary focus of this study is to verify the validity of AI driven models and their ability to enhance accuracy and overall portfolio reliability within such limitations. I evaluated two models that measure credit risk using a dataset of 10,000 loan records, testing their performance on an identical sample of 2,000 loans, by comparing six key performance indicators. The results indicate remarkable improvement in the AI model (proven by an increase of 11.6 percentage points in classification accuracy, a 22.5 percentage increase in sensitivity, an increase of 0.176 in AUC-ROC, and a reduction of 22.5 percentage points in false approval rates. By capturing hidden risk interactions, lowering Expected Credit Loss provisions and maintaining compliance under IFRS 9 and US GAAP frameworks, machine learning can considerably influence credit evaluation . But historical hesitations regarding model transparency and regulatory tension remain areas of concern that require tools like SHAP in order to be adopted into the industry.
Keywords
Credit Risk Assessment, Credit Scoring, Expected Credit Loss (ecl), Explainable Ai (shap), Ifrs 9 Compliance, Machine Learning, Random Forest