Optimal balancing & efficient feature ranking approach to minimize credit risk

Abstract

The banking industries are struggling with massive growth in the Non-Performing Assets (NPAs) that is raising the concerns of the financial institutions across the world. For gaining sustainable competitive advantages: detection, prediction, and prevention of credit Risks are becoming the foremost priorities for the banks. This data is vast, highly unstructured and imbalanced; thus, optimal balancing and efficient feature ranking are required, to predict the Credit Risk customers using Machine Learning techniques. Further, feature ranking algorithms are applied to identify the most vital characteristics of triggering the Credit Risk. The experiments have been conducted on credit Risk data set from a German bank, downloaded from the standard data repository of the UCI. Random Forest at optimal balancing ratio of 1:1.1335 has been found to be the best performing with a sensitivity of 81.6%, specificity value of 85.3%, the accuracy of 83.4%, MCC of 0.669 and AUC of 0.914.

Keywords

• Credit risk analytics
• NPA's
• SMAC
• Random forest
• Risk prediction and prevention
• SMOTE