Mining weighted-frequent-regular itemsets from transactional database

Frequent-regular itemsets mining has been explored and proposed to find interesting itemsets based on their own occurrence behavior. Traditionally, an itemset is identified as interesting, if it occurs frequently and regularly in a database. However, this task only considers items without defining difference or significance of each item which may affect the missing of important/interesting knowledge in real-world applications. To address this issue, we introduce an approach on mining weighted-frequent-regular itemsets, (also called mining WFRIs). To mine WFRIs, a tree-and-pattern growth based algorithm called WFRIM (Weighted-Frequent-Regular Itemsets Miner) is proposed. An FP-tree like structure named WFRI-tree is designed to efficiently maintain candidate itemsets during mining process. The concept of overestimated-weighted-frequency of items/itemsets under global/local maximum weight is also applied to early prune search space. Experimental results on synthetic and real datasets show efficiency of WFRIM in the terms of computational time, memory consumption and capability to find valuable itemsets.