Faculté Polytechnique De Mons

Datamining algorithms, especially those used for unsupervised learning, generate a large number of rules. It is hence impossible for an expert of the field being mined to evaluate these rules. To help carrying out the task, many interestingness measures for rules have been developed in order to filter and automatically sort a set of rules with respect to given goals. Since every measure gives a different result, and every expert has a different understanding of what a ”good” rule is, we propose in this article a double-step solution to the choice problem of a user-adapted interestingness measure: First, a characterisation of various interestingness measures, based on meaningful classical properties, is provided. Then a multiple criteria decision assistance process is applied to this characterisation and illustrates the benefit for a user, who is not an expert in data mining.