Mining maximal patterns based on improved FP-tree and array technique

Mining frequent itemsets is very important for mining association rules. However, because of the inherent complexity, mining complete frequent patterns from a dense database could be impractical, and the quantity of the mined patterns is usually very large. It is hard to understand and make use of them. Maximal frequent patterns contain and compress all frequent patterns, and the memory needed for saving them is much smaller than that needed for saving complete patterns. Thus it is greatly valuable to mine maximal frequent patterns. In this paper, the structure of a traditional FP-tree is improved and an efficient algorithm for mining maximal frequent patterns based on improved FP-tree and array technique, called IAFP-max, is presented. By introducing the concept of postfix sub-tree, the presented algorithm needn´t generate the candidate of maximal frequent patterns in mining process and therefore greatly reduces the memory consume, and it also uses an array-based technique to reduce the traverse time to the improved FP-tree. The experimental evaluation shows that this algorithm outperforms most exiting algorithms MAFIA, GenMax and FPmax.