An Algorithm for Mining Fixed-Length High Utility Itemsets

High utility pattern/itemset mining is a hotspot of data mining. Different from the traditional frequent pattern, high utility pattern takes into consideration not only the number of items in the transaction, but also the weight of these items, such as profit and price. Hence the computational complexity of this mining algorithm is higher than the traditional frequent pattern mining. Thus, one essential topic of this field is to reduce the search space and improve the mining efficiency. Constraint on pattern length can effectively reduce algorithm search space while fulfill a certain kind of actual requirement. Addressing fixed length high utility pattern mining, we propose a novel algorithm, called HUIK (High Utility Itemsets with K-lengthMiner), that first compresses transaction data into a tree, then recursively searches high utility patterns with designated length using a pattern growth approach. An effective pruning strategy is also proposed to reduce the number of candidate items on the compressed tree, to further reduce the search space and improve algorithm efficiency. The performance of the algorithm HUIK is verified on six classical datasets. Experimental results verify that the proposed algorithm has a significant improvement in time efficiency, especially for long datasets and dense datasets.