Research On Discovering Strategies Of Association Rules With A Quantitative Attribute As The Consequence

In the field of data mining,association rule is considerable important techniques. Furthmore,Apriori is the most classical algorithm of association rule.Research on the association rule begins with focusing on boolean attribute,and then nominal attribute. However,association rule regard to quantitative attribute has become one of research hotspot in this field.In this paper,firstly,we introduce the concept of association rule regard to nominal attribute,the evaluation criteria and other interrelated techniques.Then the concept of association rule regard to quantitative attribute.Following that,we introduce the equal-depth,equal-width and clustering method for discretization and analyze those methods for comparsion.Secondly,we focus on the description of the implement of Apriori algorithm and SimpleKMeans algorithm in Weka enviroment to prepare for the experiments of the algorithms.The research object in this paper is association rule regard to qantitative attribute as the consequence.Aiming at this kind of association rule, this paper introduces the process from two directions:from presupposition to consequence and from consequence to presupposition.In the first directions,we use the method for discover frequent items in Apriori algorithm to find the presupposition firstly, then combine the clustering algorithm dynamicly to find consequence according to minmum support principle.In the second directions,we firstly use three discretization methods to discrete the quantitative attributes and combine the discreted intervals dynamicly according support in order to find presuppositions.Then find the corresponding consequences in the instances sets.Finally,we conduct lots of experiments,compare the result of ARClusterer and three dynamic discretization method above to the result of Apriori' static discretization,and analyze them.The experiments shows that the second direction,achieve the dynamic change of discretization intervals according to support.It can discover some potential information that AprioriStatic algorithm may miss.However,it doesn't realize the intersection of different intervals of consequences.ARClusterer realizes both the dynamic change of discretization intervals and the intersection of intervals of consequences.