An Association Rule Miner with Semi-Autonomic Threshold Setting

Association rule mining is well-known to depend heavily on a support threshold parameter, and on one or more thresholds for intensity of implication; among these measures, confidence is most often used and, sometimes, related alternatives such as lift, leverage, improvement, or all-confidence are employed, either separately or jointly with confidence. We describe here an association mining system which requires the user to set a single parameter, of quite clear intuitive semantics, and then uses the given value to compute autonomously the corresponding thresholds for support, confidence, blocking factor (which is a slight reformulation of improvement) and confidence width (which is a complementary, recently introduced measure of novelty for association rules). We argue that the availability of one parameter is desirable; suggest a number of desiderata for the conceptual basis of such a parameter, and explain how our implementation meets them and is able to find meaningful association rules with lesser needs of domain intuition from the user.