Using Temporal-Difference Reinforcement Learning to Improve Decision-Theoretic Utilities for Diagnosis

Probability theory represents and manipulates uncertainties, but cannot tell us how to behave. For that we need utility theory which assigns values to the usefulness of different states, and decision theory which concerns optimal rational decisions. There are many methods for probability modeling, but few for learning utility and decision models. We use reinforcement learning to find the optimal sequence of questions in a diagnosis situation while maintaining a high accuracy. Automated diagnosis on a heart-disease domain is used to demonstrate that temporal-difference learning can improve diagnosis. On the Cleveland heart-disease database our results are better than those reported from all previous methods.