Measuring attractiveness of rules from the viewpoint of knowledge representation, prediction and efficiency of intervention

Rules mined from a data set represent knowledge patterns relating premises and decisions in 'if ..., then ... ' statements. Premise is a conjunction of elementary conditions relative to independent variables and decision is a conclusion relative to dependent variables. Given a set of rules, it is interesting to rank them with respect to some attractiveness measures. In this paper, we are considering rule attractiveness measures related to three semantics: knowledge representation, prediction and efficiency of intervention based on a rule. Analysis of existing measures leads us to a conclusion that the best suited measures for the above semantics are: support and certainty, a Bayesian confirmation measure, and two measures related to efficiency of intervention, respectively. These five measures induce a partial order in the set of rules. For building a strategy of intervention, we propose rules discovered using the Dominance-based Rough Set Approach - the "at least" type rules indicate opportunities for improving assignment of objects, and the "at most" type rules indicate threats for deteriorating assignment of objects. © Springer-Verlag Berlin Heidelberg 2005.