GLFMiner: Global and local frequent pattern mining with temporal intervals

Abstract

In this paper, we propose a new approach of mining temporal association rules. In conventional association rule mining algorithms, if the value of minimum support is set too high, we may lose lots of valuable rules. But if it is set too low, many trivial rules will be mined, and it is hard to distinguish which ones are valuable. When taking temporal factors into consideration, an itemset may not be frequent over the entire database but may be frequent in some specific intervals. Here, we propose a temporal association rule mining algorithm for interval frequent patterns, called GLFMiner, which can automatically generate all of the intervals without using any domain knowledge. In our algorithm, we consider not only global frequent patterns but also local frequent patterns. Then, with the same value of minimum support, we can find plenty of valuable temporal rules and don't lose any rule that conventional association rule mining algorithm can find. The experimental results show that our algorithm can mine more temporal frequent patterns than the conventional association rule mining algorithm.