Optimization Algorithm Of Weighted Association Rules Mining

As a main branch in the field of data mining, association analysis is used to find meaningful connections hidden in the data. In this area, compared with the traditional algorithm,the algorithm of weighted association rules mining is more conducive to solve the problems of the different importance and unbalance of individual items in database. Therefore, people pay more and more attentions to the related research.This paper focuses on the algorithm of weighted association rules. First, the paper introduces the popular model of weighted association rule systematically, and compares the advantages and disadvantages of them. Based on the studying the advantages of each model, I adoptd a mining strategy based on clustering and compression matrix, to mine weighted frequent itemsets, inspired by mixed weighted association rules. Its main idea is to use the matrix vector as the database data storage structure, to reduce the frequency of database access by using the method of trading space for time, to divide database by using the method of clustering partition, digging for distribution to reduce the memory footprint. Local frequent itemsets merge into global frequent itemsets. The introduction of the concept of transaction weight and database weight optimize the method of weight calculation, making it meet the monotonicity of support measures, enhancing the level of splicing and pruning. Compared with the traditional weighted algorithm, the strategy reduce the frequency of database access, improve the efficiency of the pruning candidate itemsets, and improve the overall performance and accuracy.At the same time, in order to avoid the problem that the low degree of interest, invalid and redundancy rules are dug out by single support mining rules, learning the ideas of dynamic update, this paper design a new kind of weighted association rules mining algorithm based on multiple support(A New Algorithm of Weighted Association Rules Mining with Multiple Minimum Supports, NAWARM_MMS),by the introduction the concept of Minimum support,and the combination of vertical equivalence partitioning data and itemsets. In the Algorithm, different item set corresponds to a given minimum support threshold. We mine the association rules in the different important degree of itemsets, which is small, but is more interesting, and more valuable, through to the support threshold for different projects.By testing the run time of the algorithm in different number of transactions, different support and under the condition of different number and different transactions consistency, the simulation experiment verifies that the improved algorithm of weighted association rules based on clustering and compression matrix, has more advantages, in dealing with a low degree of dense data. At the same time, the algorithm in time and space complexity is superior to the basic algorithm of weighted association rules. Further, for the improved NAWARM_MMS algorithm, test the running time in the case of different threshold levels and the different number of transactions. The results of experiment show that due to introduction of the vertical structure of database storage and itemsets equivalence partitioning, the method of I/O loading and scale of candidate itemsets generated are optimized. The overall performance is slightly better than the traditional algorithm of multiple support.