Analysis of monotonicity properties of new normalized rule interestingness measures
Roman Słowiń, Salvatore Greco, Izabela Szczęch · 2008
The paper considers interestingness measures for evaluation of relevance and usefulness of “if..., then...” rules induced from data. We propose a way to normalize three popular measures: rule interest function of Piatetsky-Shapiro, gain measure of Fukuda et al. and dependency factor used by Popper and Pawlak. The normalization transforms the measures to the interval [−1, 1], whose bounds correspond to maximal Bayesian confirmation and disconfirmation, respectively, and thus make them more meaningful. The new normalized measures are analyzed with respect to a valuable property M of monotonic dependency on the number of objects in the dataset satisfying or not the premise or the conclusion of the rule. The obtained results have a practical application as they lead to efficiency gains while searching for the best rules.