Constructing projection frequent pattern tree for efficient mining

Frequent Pattern mining plays an essential role in data mining. Most of the previous studies adopt an Apriori-like candidate set generation-and-test approach. However, candidate set generation is still costly, especially when there exist prolific patterns and/or long patterns. We introduce a novel frequent pattern growth (FP-growth) method, which is efficient and scalable for mining both long and short frequent patterns without candidate generation. And build a new projection frequent pattern tree (PFP-tree) algorithm, which not only heirs all the advantages in the FP-growth method, but also avoids it's bottleneck in database size dependence when constructing the frequent pattern tree (FP-tree). Efficiency of mining is achieved by introducing the projection technique, which avoid serial scan each frequent item in the database, the cost is mainly related to the depth of the tree, namely the number of frequent items of the longest transaction in the database, not the sum of all the frequent items in the database, which hugely shortens the time of tree-construction. Our performance study shows that the PFP-tree method is efficient and scalable for mining large databases or data warehouses, and is even about an order of magnitude faster than the FP-growth method.