Theory And Practice Of AP Algorithm And Chameleon Clustering

Data mining is a dynamic process of extracting useful information from large amounts ofdata. Clustering analysis is one of data mining methods, it does not need to make anyassumption about the number of clusters, it belongs to unsupervised learning techniques. Thesimilar data objects are classified into the same cluster, and different data objects intodifferent clusters.Hierarchical Chameleon algorithm constructs similarity function based on the relativeconnectivity and relatively closeness between different clusters, and maximize the functionvalue to implement clustering. Chameleon algorithm can discover clusters of arbitrary shapeswith high quality, but it must be given the value of k which is used to construct k-nearestneighbor graph, and it must select the threshold of similarity function and the value of thesmallest bisection. Therefore, we propose an improved Chameleon algorithm (GMChameleon)to solve the drawbacks of the Chameleon algorithm. But for high dimensional dataset, theGMChameleon algorithm has poor results and requires more memory support.In this thesis, we propose a novel algorithm named as PAPGMChameleon to improve theclustering effect using the Affinity Propagation clustering algorithm combined with theGMChameleon algorithm. We use the Affinity Propagation algorithm for the preprocessingphase, the dataset is divided into subsets of data. Then the GMChameleon algorithm are usedfor further clustering. The experiment results on five dataset indicate the surperiority of thePAPGMChameleon algorithm. The performance of the PAPGMChameleon algorithm iscompared with the AP, Chameleon, GMChameleon and APSCAN algorithm.