Enhancing Fairness and Accuracy in Credit Score Analysis: A Novel Framework Utilizing Kernel PCA

The credit score (CS) analysis is a crucial tool in the financial sector, offering a standardized method to evaluate the risk associated with the credit and assist in making informed lending decisions. It creates transparency and promotes responsible financial behavior by empowering the lender and borrower. The likelihood of repaying debts by the borrower is analyzed based on the credit history and financial behavior and lenders make decisions about extending credit. Higher credit scores offer more favorable loan terms, lessen the rate of interest, and enhance borrowing capacity. However, a lower credit score augments the interest rates or complexities in obtaining credit. Various automated CS analysis techniques developed poses multiple challenges including credit bias and fairness challenges, complexity in handling data, and inaccurate prediction score. To conquer the above-mentioned drawbacks, this study develops a novel methodology for enhancing fairness, transparency, and accuracy in credit scoring. Here the input data collected from the credit score prediction dataset is preprocessed initially. Then the original data is transferred as numerical features with the KPCA technique. KPCA extracts the non-linear associations among the data and reduces the feature space dimensionality. Further, the credit score is predicted with the attention-based RNN which handles input sequences of varying lengths, accommodating individuals with different credit histories. Empirical results indicate that the developed model improves the prediction performance and mitigates possible economic losses significantly while assessing credit risks.