Analysis in HUGIN of Data Conflict

Abstract

After a brief intro duction to causal proba­ bilistic networks and the HUG IN approach, the problem of conflicting data is discussed. A measure of conflict is defined, and it is used in the medical diagnostic system MUNIN. Finally it is discussed how to dis­ tinguish between conflicting data and a rare case. quacy of the model and the reliability of data used. Therefore, no expert will blindly accept what the system comes up with. At least there will be kept a critical eye on the data, and mainly one will look for conflicts in the data or conflicts with the model. In this paper we present a way of building such a critical eye into a system with a CPN model. Our suggestion requires an easy way of calculating probabilities for specific configurations. We start with a brief introduction to the HUGIN approach. In section 3 we discuss CPN's and data conflict. In section 4 a measure of conflict is defined, and it is shown that this measure is easy to calculate in HUGIN and that it supports a decomposition of global conflict into local conflicts .. Section 5 reports on experience with a large CPN, and in section 6 we discuss how to distinguish between conflicts in data and data originating from a rare case.