Multi-level Frequent Pattern Mining

Abstract

Frequent pattern mining (FPM) has become one of the most popular data mining approaches for the analysis of purchasing patterns. Methods such as Apriori and FP-growth have been shown to work efficiently in this setting. However, these techniques are typically restricted to a single concept level. Since typical business databases support hierarchies that represent the relationships amongst many different concept levels, it is important that we extend our focus to discover frequent patterns in multi-level environments. Unfortunately, little attention has been paid to this research area. In this paper, we present two novel algorithms that efficiently discover multi-level frequent patterns. Adopting either a top-down or bottom-up approach, our algorithms exploit existing fp-tree structures, rather than excessively scanning the raw data set multiple times, as might be done with a naive implementation. In addition, we also introduce an algorithm to mine cross-level frequent patterns. Experimental results have shown that our new algorithms maintain their performance advantage across a broad spectrum of test environments.