The Research On The Algorithm Of Mining Quantitative Association Rules

Nowadays, data which contains much important information is increasing quickly. It isarduous task to mine valuable information or knowledge. Data mining is developing quicklyto meet this demand. Data mining has been defined as"The nontrivial extraction of implicit,previously unknown, and potentially useful information from data in large database or datawarehouse". Mining association rules in large relational databases is an important researchtheme in data mining.Association rules can be classified into boolean association rules and quantitativeassociation rules by the objects which will be handled. Mining boolean association rules is tofind associations between attributes whose values are all"1"in relational table where everyattribute is boolean. In real life, the attributes in relational databases are normally quantitative.Therefore, it is very significative to study on how to mine quantitative association rules.In this paper, we have done some researches on the problems of mining quantitativeassociation rules. It is stated as follows:A novel algorithm MQAR is proposed to mine quantitative association rules. Thealgrithm combines FP-tree in mining frequent patterns and CLIQUE which is used forclustering, and we design a new data structure named DGFP-tree to save the information ofthe database and subspaces which have clusters. Then mining quantitative association rules istransformed into the problem of constructing DGFP-tree and clustering by searching denseunits in the DGFP-tree. The algorithm not only can avoid the conflict between minimumsupport problem and minimum confidence problem, but also can mine important associationswhich may be missed by previous algorithms. Experimental results show that MQAR canefficiently find quantitative association rulesAgainst the problems of discretizations and combinations of quantitative attributes, we propose a novel algorithm FMI-Miner which is based on fuzzy clustering and mutualinformation to mine quantitative association rules. The algorithm firstly partitions attributesby fuzzy C-mean clustering, then mines frequent fuzzy itemsets by mutual informationentropy to form association rules. The experimental result shows that the mining efficiency ofFMI-Miner has been improved and the association rules mined are more understandable.