F-Miner: A New Frequent Itemsets Mining Algorithm

Abstract

In this paper, we present a novel algorithm, called F-Miner, to mine the complete set of frequent itemsets by pattern growth. The F-Miner algorithm uses two new compact data structures, ascending FP-tree (AFP-Tree) and frequent pattern forest (FP-forest), to represent the conditional databases. When we construct an AFP-tree, the items infrequent 1-itemset are ordered in frequency ascending order. The AFP-Tree structure is traversed in top-down depth-first order. The root of the AFP-Tree is not "null", but an item which can identify this tree. AFP-tree has a one-dimensional array which stores the counts of every tree-node's item except root-node. In F-Miner, we need many AFP-trees to store a conditional database; these trees construct one forest, called FP-forest. We test our algorithm versus several other algorithms on real world datasets, such as BMS-POS. The experimental results show that our algorithm is an efficient algorithm on both sparse and dense databases