Fast algorithm for mining multilevel association rules

Abstract

We present two algorithms, AprioriNewMulti and AprioriNewSingle, for data mining multilevel and single level association rules, respectively, in large databases. The database consists of following fields, transaction ID and items purchased in the transaction. The algorithms introduce a new concept called multi minimum support i.e. minimum support varying for different lengths of the item set. Unlike other algorithms, AprioriNewMulti does not depend on the number of levels in the concept hierarchy, i.e., it does not scan the database for each level of abstraction for finding association rules. Scale up experiments show that both of these algorithms have scale linear with the number of customer transactions.