Ensembling of Performance Metrics in Credit Risk Assessment Using Machine Learning Analytics

A major financial risk to commercial banks is credit risk, which arises when borrowers default on their debt. Accurately predicting the credit risk of their clients is a key difficulty faced by these institutions. Since these institutions depend entirely on these prediction models, it is essential that the performance level of the model be high. Depending solely on accuracy can be risky due to the inherent nature of imbalance in most datasets. Traditional statistical methods are insufficient for addressing this imbalance issue. In order to identify the issues related to imbalance problems, it is essential to employ a sufficient number of evaluation metrics. Hence, the evaluation process encompassed nine performance parameters, namely Accuracy (ACC), Specificity (SPEC), Matthews Correlation Coefficient (MCC), Cohen Kappa Score (CK), F1-Score, Precision (PRE), Recall (REC), G-mean, and RoC Score. In addition, a custom metric ‘Pmean’ has been suggested and validated using fourteen credit risk datasets. The study's exhaustive methodology seeks to not only find out the best model in terms of the performance but also guarantee the true nature of the model in the context of credit risk evaluation. The research provides new insights into the field of credit risk prediction by conducting a thorough analysis using twelve classification approaches and machine learning models.