Statistically Significant Pattern Mining With Ordinal Utility

Statistically significant pattern mining (SSPM), which evaluates each via a hypothesis test, is an essential and challenging data task for knowledge discovery. We introduce preference relation between patterns aim to discover the most preferred under constraint of statistical significance, has never been considered in existing SSPM problems. propose iterative multiple testing procedure that can alternately reject safely ignore less useful hypotheses than rejected one. By filtering out with low utility, we avoid significance budget consumption rejecting useless (uninteresting) focus on more patterns, leading discoveries. show proposed method control familywise error rate (FWER) certain assumptions, be satisfied by realistic problem class SSPM. also always discovers equally or Tarone-Bonferroni Subfamily-wise Multiple Testing (SMT). Finally, conducted several experiments both synthetic real-world evaluate performance our method. The discovered many datasets all five tasks.