Parallel Method for Discovering Frequent Itemsets Using Weighted Tree Approach

Every element of the transaction in a transaction database may contain the components such as item number, quantity, cost of the item bought and some other relevant information of the customer. Most of the association rules mining algorithms to discover frequent itemsets do not consider the components such as quantity, cost etc. In a large database it is possible that even if the itemset appears in a very few transactions, it may be purchased in a large quantity. Further, this may lead to very high profit. Therefore these components are the most important information and without which it may cause the lose of information. This motivated us to propose a parallel algorithm to discover all frequent itemsets based on the quantity of the item bought in a single scan of the database. This method achieves its efficiency by applying two new ideas. Firstly, transaction database is converted into an abstraction called weighted tree that prevents multiple scanning of the database during the mining phase. This data structure is replicated among the parallel nodes. Secondly, for each frequent item assigned to a parallel node, an item tree is constructed and frequent itemsets are mined from this tree based on weighted minimum support.