Is there a way to remove redundant rules produced by Association Rule Learner, perhaps via subset matcher?
Thanks in advance.
Sajjad
I'm looking for that same capability. Obviously if the support and confidence are identical, they are likely redundant (I usually opt for the shortest rule in this case, i.e., smallest # of antecedents).
But I haven't worked out a way to do this yet in KNIME without resorting to a java snippet node to access prior records in the itemset. (I might sort by Confidence, descending, then support, then itemset length ascending and keep the first of any group of rules).
Better still would be to remove nearly identical rules, where nearly can be defined in any number of ways. one way is with a chi-square test for the pair of rules being compared. But I'd have to build that out with a loop.
Looping actually makes me wonder.... I can imagine removing redundant rules with a loop over rows and doing a test: if the confidence and support are different from the prior values, keep the rule, otherwise dump it.
I'll let you know if am successful (presuming I have time to try it)