Navigating Class Disparity: A Systematic Literature Review of Credit Risk Prediction

Purpose: The purpose of this paper is to systematically review of the current research on the topic of credit risk prediction. The aim is to identify the models commonly used for predicting credit risk and to understand the types of variables incorporated into these models. The review will also explore the methodologies that have been employed to tackle the issue of class imbalance, a significant challenge that can affect the performance and reliability of prediction models. Furthermore, this study is aimed to examine the different approaches used to interpret the models' outcomes. This will contribute significant insights to the academic community, financial professionals, and policymakers. Design/methodology/approach: The PRISMA approach is used to guide the review process. Findings: The results of the systematic review present a variety of models for predicting credit risk, with the CatBoost and Random Forests models combining to exhibit the best results. The findings underline the challenge of dealing with imbalanced datasets in machine learning and present a number of solutions. The current review highlights that one-class classifiers might be the most effective way to address class imbalance. The study additionally emphasizes how crucial interpretability is to enhance the understanding of prediction and improve model performance. Originality/value: To the authors' best knowledge, this is likely the first work that suggests the use of one-class classifiers to address the issue of class imbalance in data derived from real-life scenarios like credit risk and fraudulent activity.