Efficiently finding high utility-frequent itemsets using cutoff and suffix utility

Abstract

High utility itemset mining is an important model with many real-world applications. But the popular adoption and successful industrial application of this model has been hindered by the following two limitations: (i) computational expensiveness of the model and (ii) infrequent itemsets may be output as high utility itemsets. This paper makes an effort to address these two limitations. A generic high utility-frequent itemset model is introduced to find all itemsets in the data that satisfy user-specified minimum support and minimum utility constraints. Two new pruning measures, named cutoff utility and suffix utility, are introduced to reduce the computational cost of finding the desired itemsets. A single phase fast algorithm, called High Utility Frequent Itemset Miner (HU-FIMi), is introduced to discover the itemsets efficiently. Experimental results demonstrate that the proposed algorithm is efficient.