Research On Clustering Algorithms Using Kernel-Based Learning Method

The data mining is a multi-disciplinary overlapping research area, which is closely related with the machine learning and statistics. As a main idea of data mining, the cluster is the process of gathering a set of physical or abstract object into classes of similar object. As this method with "no supervision" has important applications in many ways, such as machine learning, pattern recognize and so on.In recent years, with the consummation of the statistics theory of learning's, the kernel-based learning methods have been proposed constantly. On the base of related knowledge of cluster, this paper deals with the cluster question in the data mining, using the kernel-based learning methods, and the kernel-based K-means clustering algorithm and the support vector clustering algorithm has been mainly studied and analyzed.Kernel-based K-means clustering algorithm is firstly map the data from their original space to a high dimensional space then perform K-means clustering in the high dimensional space. The data are exported to be more separable through the Kernel function mapping, thus realizes a more accurate cluster.Support vector machines (SVM) were a new machine learning method based on the statistics theory of learning (STL) in the mid-1990s. It the biggest gap principle and the nuclear function theory unify in together has solved much difficult problems, such as the small sample, the high dimension. And support vector clustering is a novel clustering method inspired by support vector machines and kernel study method. With comparing with the other tradition clustering methods, we can find that the parameters of SVC are comparatively few and SVC can easily deal with high dimensional data have lots of advantages, can yield a global optimum; and it can treat arbitrary shape data set, and needn't specify the number of the clusters ahead.The main contents of this paper are as follows:(1) Study on the hierarchical clustering algorithm, K-means clustering and self-organization competition three kinds of classical clustering algorithms.(2) Based on the classic Kernel-based K-means clustering algorithm and Support Vector Machine Theory and kernel-based learning methods, Study on kernel-based K-means clustering algorithm and support vector clustering algorithm, and gives an improved support vector clustering algorithm.(3) To verify the effectiveness of the algorithm, the use of artificial data sets and standard data sets to study the performance of the Kernel-based K-means clustering algorithm and support vector clustering algorithm. Through algorithm performance analysis with the different parameter and compares experiment with the classical clustering algorithms, the results show that the above-mentioned clustering algorithm kernel-based learning method has good stability, and has the significant advantage of an ideal clustering effect.