A Clustering of Interestingness Measures

摘要: It is a common issue that KDD processes may generate large number of patterns depending on the algorithm used, and its parameters. hence impossible for an expert to sustain these patterns. This be case with well-known Apriori algorithm. One methods used cope such amount output depends use interestingness measures. Stating selecting interesting rules also means using adapted measure, we present experimental study behaviour 20 measures 10 datasets. compared previous analysis formal meaningful properties measures, by two clusterings. goals this enhance our approach. Both approaches seem complementary could profitable problem user’s choice measure.