Mining informative rule set for prediction over a sliding window

Abstract

We study the problem of mining informative (association) rule set for prediction over data streams. On dense datasets and low minimum support threshold, the generating of informative rule set does not use all mined frequent itemsets (FIs). Therefore, we will waste a portion of FIs if we run existing algorithms for finding FIs from data streams as the first stage to mine informative rule set. We propose an algorithm for mining informative rule set directly from data streams over a sliding window. Our experiments show that our algorithm not only attains high accurate results but also out performs the two-stage process, find FIs and then generate rules, of mining informative rule set. © 2010 Springer-Verlag Berlin Heidelberg.