Gregory Piatetsky-Shapiro and Christopher J. Matheus
One of the moet promising areas in Knowledge Discovery in Databases is the automatic analysis of changes and deviations. Several systems have recently been developed for this task. Suc~ of these systems hinges on their ability to identify s few important and relevant deviations among the multitude of potentially interesting events:~ In this paper we argue that related deviations should be grouped togetherin a finding and that the interestingness of a finding is the estimated benefit from a poesible ~tion connected to it. We discuss methods for determining the estimated benefit from the impact of the deviations and the success probability of an action. Our analysis is done in the context of the Key Findings Reporter (KEFIIt), a system for discovering and explaining "key findings" in large relational databases, currently being applied to the analysis of healthcare information.