The Research And Implementation Of Density-based Clustering Algorithm With Pattern Evaluation Methods

As a hotspot of computational domain, Data Mining technology and its subdiscipline clustering analysis attracts lots of researchers'prevenance. A large number of clustering algorithms have been developed in a variety of domains for different types of applications, but none of these algorithms is suitable for all types of data, clusters, and applications.In fact, it seems that there is always room for a new clustering algorithm that is more efficient or better suited to a particular type of data, cluster, or application. In the meantime, clustering pattern evaluation technology which is expected to judge what constitutes a good set of cluster has been developed. However, when an objective measure is employed to give a precise definition of a cluster, the problem of finding the optimal clustering is often computationally infeasible.This paper analyzes clustering algorithms particularly at first, and then it researches the two aspect of the study task in detail, the amelioration about density.based clustering algorithm and exploration new cluster evaluation method aided with large numbers of scatters and experimental consequential tables. Because of the problem of classical algorithms, such as ascertainment about initial qualification, input parameter'dependence upon domain knowledge, infection from outlier data, various density data clustering, and so on, this paper presents a new algorithm:Clustering Algorithm Based on Density and Density-reachable (CADD). The innovation of CADD algorithm goes as follows: 1.The concept of density.reachable indirectly. 2. The concept of local density. 3. The dynamic neighborhood radius. The result of experimentation makes it evidence that CADD algorithm is successful in arbitrary shape cluster, and various density data.The second work of this artical is the exploration about clustering patten evaluation. The experimentation result about this paper'new pattern evaluation which can be used to evaluate individual clusters and objects based on the k-nearest neighbor graph of data demonstrates that this method is practically feasible and enhances the ability of explanation and implementation of clustering results.At the last phase, the paper validates the effect of an actuality dataset used CADD clustering algorithms with patten evaluation compared of K-means, Hierarchical clustering and neural network method.