Associative Classification Based On Hybrid Strategy

With the rapid growth of information data, data mining has become an activeresearch field, which is used to obtain useful information from massive data quickly andefficiently. Association rule mining and classification mining are two importantbranches in the field of data mining, their applications spread in various fields.Associative classfication method is a new method based on the combination ofassociation rule mining and classification mining, which has the advantages of highclassification accuracy and expansibility etc. and has attracted the attention of manyresearchers. It is a research direction which has great research value and applicationprospect in the field of classification.The existing associative classification methods mainly include Eager method andLazy method. Eager associative classification method is an overall analysis carried outfor the entire samples, but there is small disjunction problem when analyzing a smallamount of class samples. If by reducing the threshold of minimum support andminimum confidence to solve the problem, it will lead to greater mining cost. However,Lazy associative classification method can avoid small disjunction problem accordingto analyze the specific sample. But when there are a lot of samples to be classified, itwill appear the problem of low efficency of classification.Firstly, this thesis introduces the related theories and algorithms of association rulemining, then analyzes the advantages and disadvantages of Eager method and Lazymethod. Secondly, according to the deficiency of the two algorithms, the thesis putsforward an new classifiation method which is associative classification based on hybridstrategy by combining their advantages, and analyzes the classification of missing datasets. Finally, the experimental results on18datasets from UCI Machine LearningRepository verify the effectiveness of this method. The main contributions of theproposed method in this paper are as follows:â‘ This paper puts forward a new model of associative classification based onhybrid strategy combining Eager method and Lazy method, which aimed at thedisadvantages of two kinds of assoviative classification methods.â‘¡In the aspect of algorithm, using Lazy method to solve small disjunctionproblem of Eager method, and in this paper we improve the Eager method and rulematching. The experimental results show that associative classification based on hybrid strategy compared with CBA method and LAC method has the highest accuracy.â‘¢This method is greatly improved compared with Lazy method in the executionefficiency, and it makes associative classification method more practical.â‘£In order to verify the classification of this method on the missing data sets,experiments are conducted with the natural missing data sets and artificial missing datasets which making the deletion processing of5%,10%,15%,20%. The experimentalresults show that the method has higher accuracy than NaiveByes when directlyclassifies the missing data sets.