The Research Of Data Mining Methods Based On Association Rules

Data mining is a hotspot in the fields of artificial intelligence and database technology, and is playing a powerful role in the practical application. Association rules mining is the most active and important research topic in data mining, aims at discovering the interesting associations and correlations between items in the given dataset. Algorithms of traditional association rules treat each item and record uniformly. In fact, users often distinguish the importance of the items and records in order to find more interesting and valuable rules. Weighted association rules mining can solve the problem and is drawing more and more domestic and foreign researchers'attention.In this paper, the related knowledge of data mining and association rules is outlined. Then, vertical, horizontal and mixed weighted association rules are introduced systematically. The common models of weighted association rules are also discussed and analyzed. The weighted association rules are studied intensively and the improved algorithms and the applications are proposed.Firstly, The defects of the New_Aprior are pointed out, and the shortages of the improved algorithm—MWFI(Mining Weighted Frequent Itemsets)are analyzed in detail. Then the New-MWFI algorithm for mining weighted association rules was proposed which reflects the items'different importance. In this algorithm the transactions are classified according to the item's weight and mining the frequent itemsets of each category satisfies the Apriori's property, so the Apriori algorithm or other improved algorithms can be used, the mining efficiency is improved. The improved algorithm also can be easily extended to the mixed weighted association rules mining.Additionally, the frequencies of the items are always different in the database, so different minimum supports are required for different items. Algorithm of mining weighted association rules with multiple minimum supports was proposed, which allows specifying different supports for different items and considering the records'different importance. In the mining, this algorithm overcomes the defect of the multiple minimum supports which is not satisfying the Apriori's property, and it needn't scan the database many times repeatedly. The redundant items are deleted and the same itemsets are cumulated, too. The experiments show the algorithm's efficiency.Finally, the weighted association rules are applied to the temporal database, and a mining algorithm of weighted temporal association rules is proposed. The new algorithm provides an effective pruning method, overcomes the shortages of using the frequent (k-1)-itemsets to generate the candidate k-itemsets directly in the existing algorithms which use item's lifespan as time feature. The example of analysis and the experiment of comparison demonstrate that the algorithm is effective. The rules not only highlight the item's weight but also reflect the temporal semantics of the real database, and would be more useful for the practical application.