Mining infrequent and interesting rules from transaction records
Alex Tze Hiang Sim, Maria Indrawan, Bala Srinivasan · International Conference on Artificial Intelligence · 2008
In association rule mining, apriori based techniques require users to define a minimum support threshold before rules are mined. These techniques suffer from two disadvantages. First, it requires a very low pre-set minimum support threshold to cover all rules that may be interesting if measured using other measures. Second, the number of rules reported is simply too large. We augment the measure of proportional error reduction, and describe a technique to mine interesting association rules directly. Our approach is unique from apriori based techniques, and do not require a minimum support threshold. Hence it is immune from its drawbacks. We also present two pruning opportunities that can be applied to the algorithm to reduce the computational cost of the mining process. Our experiments show that our approach can find proportional error reduction rules directly from databases. Key-Words: association rule mining, infrequent, proportional error reduction, rule pairs, PRE rules