Exploring conformal prediction for uncertainty quantification: an application in loan approval

Abstract

In the era of machine learning dominating high-stakes decision-making, it is essential to employ methods that not only generate predictions but also measure the associated uncertainties. This is especially important considering the profound impact such decisions can have on human lives. Depending on the context of application, relying solely on point predictions can lead to devastating consequences. Hence, the need for new techniques that measure uncertainty in predictions arises, aiming to contribute to the process of ensuring fairness and robustness in decision-making. Conformal Prediction stands out as a promising framework to address this challenge. This study outlines the mathematical foundations of Conformal Prediction as a solution for uncertainty quantification. Furthermore, it explores the implementation of this model as a business tool for loan approval, showcasing its practical application in credit risk assessment. This application demonstrates how concepts from a Mathematics bachelor’s program can be effectively integrated with knowledge from a Business Administration program to develop a valuable business tool.