Abstract

Abstract: High utilityitem set mining (HUIM) has emerged as an important research topic in datamining, with applications to retail-market data analysis, stock marketprediction, and recommender systems, etc. However, there are very few empiricalstudies that systematically compare the performance of state-of-the-art HUIMalgorithms. In this paper, system presents an experimental evaluation on HUIMalgorithms. Our experiments show that EFIM are generally the top performers inrunning time, while EFIM also consumes the least memory in most cases. In orderto compare this algorithm in depth, we use another synthetic datasets withvarying parameters so as to study the influence of the related parameters, inparticular the number of transactions, the number of distinct items and averagetransaction length, on the running time and memory consumption of EFIM. In thiswork, we demonstrate that, EFIM is more efficient under low minimum utilityvalues and with large sparse datasets, in terms of running time; although EFIMis the fastest in dense real datasets, it is among the slowest algorithms insparse datasets. We suggest that, when a dataset is very sparse or the averagetransaction length is large and running time is favoured over memoryconsumption. This work has reference value for researchers and practitionerswhen choosing the most appropriate HUIM algorithm for their specificapplications.

Keywords: High-Utility Mining, Item Set Mining,Pattern Mining.