Conformal Predictive Decision Making

Abstract

In many real-world settings, machine learning (ML) predictions serve as intermediate outputs used to inform decision-making. However, quantifying and accounting for uncertainty in these decisions remains a fundamental challenge. Conformal Predictive Decision Making (CPDM) is a framework for decision-making under uncertainty that leverages Conformal Predictive Distributions (CPDs) to optimize outcomes over a specified utility function. In this work, we evaluate two CPDM variants on synthetic datasets and compare their performance to two alternative approaches: Bayesian Decision Theory (BDT) and Point Predictive Decision Making (PPDM). Our proposed CPDM algorithm significantly outperforms the previously established one in most cases, while also offering computational advantages. It also showed greater robustness than BDT and PPDM in scenarios involving noisy data and skewed utility functions.