Research On Skyline Pattern Mining Algorithm

Itemset mining is one of the most typical research directions in the field of data mining.Most of the traditional itemset mining algorithms are limited to mining itemsets with high frequency or high utility by considering the itemsets' frequency or utility alone.Skyline frequent utility itemsets mining will consider both the frequency of the itemsets and the utility of the itemsets.Mining those itemsets that are located on the skyline and not dominated by other itemsets,namely the skyline frequent utility itemsets(SFUI),can be expressed as decision making provides richer information.This paper proposes two SFUI mining algorithms from the perspectives of branch constraint pruning and heuristic search,namely,the SFUI-UF algorithm based on utility filtering and the SFU-CE algorithm based on cross entropy.In contrast,SFUI-UF can mine all SFUIs by using traditional pruning strategy,while SFU-CE using heuristic search method can find most SFUIs with high efficiency.(1)For the SFUI-UF algorithm,first consider filtering by itemsets frequency,design a new maximum utility array,and verify that its size is not larger than the array size in the state-of-the-art algorithm,and prune hopeless itemsets and its extension.Secondly,using transaction weighted utility filtering,the minimum utility of SFUI is proposed and proved that the theory can reduce the search space in the initial stage of the algorithm.In addition,considering using the itemset utility itself for filtering,a minimum expansion utility is proposed and verified as a pruning strategy in the expansion phase of the search space traversal.Finally,extensive experimental results show that the SFUI-UF algorithm can discover all correct SFUIs with high efficiency and low memory consumption.(2)In the SFU-CE algorithm based on cross-entropy,firstly,the SFUI mining problem is modeled by using cross-entropy with the itemset utility as the optimization object.Then those itemsets with utility or utility upper bounds are pruned using critical utility of SFUI.In addition,a random mutation strategy is designed to make the itemsets in each sample more diverse.Finally,its efficiency and accuracy are demonstrated by comparing SFU-CE with state-of-the-art algorithms.Based on the above research,it can be seen from the performance evaluation that the skyline pattern mining algorithm proposed in this paper has the characteristics of high efficiency,high precision and scalability,and has certain research significance for skyline pattern mining and heuristic search algorithm research.