Self-similar mining of time association rules

Abstract

Although the task of mining association rules has received considerable attention in the literature, algorithms to find time association rules are often inadequate, by either missing rules when the time interval is arbitrarily partitioned in equal intervals or by clustering the data before the search for high-support itemsets is undertaken. We present an e.cient solution to this problem that uses the fractal dimension as an indicator of when the interval needs to be partitioned. The partitions are done with respect to every itemset in consideration, and therefore the algorithm is in a better position to find frequent itemsets that would have been missed otherwise. We present experimental evidence of the e.ciency of our algorithm both in terms of rules that would have been missed by other techniques and also in terms of its scalability with respect to the number of transactions and the number of items in the data set.