Research On Dynamic Association Rules Mining Algorithm

Association rules mining is one of the significant issues in Data Mining, which describes potential relationships among data items in database.The Apriori algorithm is one of the most frequently used algorithms. But the Apriori algorithm has some defects, for example, it will produce large amounts of candidate itemsets and need scanning whole database frequently. Many researches try to improve the performance of the Apriori algorithm, but still can't escape from the frame of the Apriori algorithm mostly and lead to a little improvement of the performance. In addition, there are two prevalent problems in association rules mining: First, usually, it's necessary to set some parameters for customers before mining, and mostly they have to adjust these parameters many times to acquire the satisfactory rules, then how to implement efficiently during the repetitious process? Second, how to acquire the desired results efficiently and immediately when the mining data updates constantly? The traditional association rules mining algorithms are static, for the above problems, they must re-process the whole database again to make sure the consistence between association rules and data, result in dissatisfactory efficiency.To solve the first problem, a new associate rules mining algorithm, named EUF, is proposed in this paper. The times that the EUF algorithm scans the transaction database needn't more than twice to calculate the support counts of all candidate itemsets by using of subset temple mapping. Then, input the support thresholds and confidence thresholds to generate the associate rules by user. The efficiency of EUF is independent of the support thresholds, and doesn't need re-process the whole database when the support threshold is adjusted. So the efficiency of EUF is steady and efficient. To solve the second problem, we propose an algorithm, named EUF-IU, based on EUF. Regardless of how the database changes, the EUF-IU only processes the modified part in the database, instead of the whole database, and then get accurateand complete association rules.Therefore, it can save the cost of runtime. Experiments show that the two algorithms presented by this paper can dynamically mine the association rules with the changing of parameters and database, and the efficiency of the two algorithms is superior to the traditional association rules mining algorithms when the minsupports are small, especially better than the QDT algorithm.