Research On Evolution Clustering Algorithm For Dynamic Data Mininag

Along with the development of information technology, data mining is applying more andmore widely, Traditional data mining usually discovers knowledge from static database, but thedata of application field is mostly dynamical, All the data in the database change over time. Whenusing clustering method to cluster data, If every time after data update, we all clustrer the data setOn the one hand the price is too big; On the other hand, because of no using the first clusterrelevant information, computing resources waste,so, Designing dynamic incremental evolutionclustering algorithm to improve the clustering efficiency becomes necessary. This paper appliedartificial immune system and fractal theory to clustering, put forward two kinds of dynamicevolution clustering algorithm aiming at the dynamic data set.The major work and innovation ofthis paper are as follows.Introducing kernel function into artificial immune clustering, this paper proposed an artificialimmune dynamic clustering algorithm based on kernel function KAIDA. KAIDA uses kernelfunction method to map data to high dimensional feature space, computes the kernel spacedistance between new antigen and existing memory antibodies, by comparing it with the memoryantibody center’s recognition radius, the algorithm determines to classify the new antigen toexisting cluster, or to form a new class for it. The experimental results show that the KAIDAalgorithm can effectively realize dynamical and self organizing clustering, besides,comparingwith the artificial immune algorithm without using kernel function, the KAIDA algorithm canreduce the data number much better that belongs to different cluster but classified together, hashigher correct classification rate. In order to avoid the phenomenon that memory antibody is toocentralized, and to improve the compression ratio of clustering results,the algorithm carries outmemory antibodies immune suppression operation for algorithm optimization, The experimentalresults show that the optimization algorithm improved the compression ratio of clustering results,increased the algorithm rationality(2)Proposed a dynamic evolution clustering algorithm based on artificial immune and fractal.The algorithm considers the sensitivity of fractal clustering for the initial clustering results,andthe high accuracy that using artificial immune nuclear clustering method for data cluster, Choosesartificial immune nuclear clustering method for data clustering to form the fractal clustering initialresults, Then, using the characteristics that partial distribution of fractal has similar structure orattribute with the whole distribution of fractal, carries out expand clustering for data set. theexperimental results show that the clustering results has higher accuracy...