Efficient Infrequent Itemset Mining Using Depth-First and Top-Down Lattice Traversal

Frequent itemset mining is substantially studied in the past decades. In varies practical applications, frequent patterns are obvious and expected, while really interesting information might hide in obscure rarity. However, existing rare pattern mining approaches are time and memory consuming due to their apriori based candidate generation step. In this paper, we propose an efficient rare pattern extraction algorithm, which is capable of extracting the complete set of rare patterns using a top-down traversal strategy. A negative item tree is employed to accelerate the mining process. Pattern growth paradigm is used and therefore avoids expensive candidate generation.