Maximum Item First Pattern Growth for Mining Frequent Patterns

Abstract

Frequent pattern mining plays an essential role in many important data mining tasks. FP-Growth, an algorithm which mines frequent patterns in the frequent pattern tree (FP-tree), is very efficient. However, it still encounters performance bottlenecks when creating conditional FP-trees recursively during the mining process. In this work, we propose a new algorithm, called Maximum-Item-First Pattern Growth (MIFPG), for mining frequent patterns. MIFPG searches the FP-tree in the depth-first, top-down manner, as opposed to the bottom-up order of FP-Growth. Its key idea is that maximum items are always considered first when the current pattern grows. In this way, no concrete realization of conditional pattern bases is needed and the major operations of mining are counting and link adjusting, which are usually inexpensive. Experiments show that, in comparison with FP-Growth, our algorithm is about three times faster and consumes less memory space; it also has good time and space scalability with the number of transactions.