Statistical pruning of discovered association rules (Q1861621)

The authors propose a new strategy which allows the user to find ``interesting'' subsets of statistical association rules in market basket analysis. To do this, some statistical tests are applied to the set of rules which evaluate (1) how much the antecedent part is able to explain the variability of the consequence, (2) how significant are the logical implications of the rule, (3) how strong are the logical implications. This strategy is applied to the analysis of a data set containing preferences for Italian national television channels.