Research On Association Rules Mining Algorithm Based On Item Reduction

In recent years, with the rapid development of information technology and the increasing amount of information, traditional statistical methods can no longer meet people's need. As an important research topic in the field of data mining, association rules reflects the interdependence or relevance between objects. The research on how to mining the association rules more efficiently is of great theoretical value and practical significance.The main task of this dissertation is the research on the relationship between association rules and the data itself. We also have achieved certain innovative contributions. The main research work of the dissertation can be summarized as the follows:Aiming at the shortcoming of current Aproiri algorithm, a new solution is proposed in the dissertation. The Apriori algorithm is a classic method for mining association rules. However when the support threshold is lower, a huge set of candidates will be generated while a large amount of time will be consumed to count their supports. According to the problem, this dissertation presents a new conception of item transaction and an Apriori algorithm based on item reduction (Apriori-IR). The new algorithm can improve the deficiency and the efficiency of Apriori algorithm by using item reduction operations. The dissertation proves that Apriori-IR can reduce the times of joining and pruning, also it can cut down the time of candidates counting. The experimental results show the feasibility and effectiveness of the algorithm.For further research on the effects of the item reduction on the association rules mining algorithm, FP-growth algorithm is used to mining the data processed by the item reduction operation, and the FP-GIR algorithm (an FP-growth algorithm based on item reduction) is proposed. It is proved that the FP-GIR algorithm can reduce the consumptions of the time space and memory space. The feasibility and effectiveness of the algorithm can be shown in the experimental results on different datasets. A FP-treeIR algorithm based on FP-tree structure is put forward at last. Also the experiments show that the algorithm can reduce the consumption of the time space.