Research Of Support Vector Clustering Algorithm Based On Weighted Gaussian Kernel With Multiple Widths

The method of machine learning to analyze data and excavate the information behind mass data has facilitated the production of data mining. Cluster analysis has significant function for data mining. As a data partitioning, it can make data with same properties delivered unto the same class and distinguish exceptional data point. Support Vector Machine (SVM) as a technique of data mining proposed by Vapnik in 1990s is a new tool for machine learning recur to optimization method. SVM works in feature space by kernel mapping, and this aims to acquire linear property in feature space for nonlinear problems in input space. Kernel function, as an important approach to realize nonlinear mapping, is the key to the widespread and successful application of Support Vector Machine. Support Vector Clustering(SVC)is a novel clustering method using the approach of SVM, while Gaussian kernel with single width generates underfitting learning for sparse areas and overfitting learning for dense areas, by which the generalization of SVC is limited. So a new kernel function called Weighted Gaussian Kernel with Multiple Widths(WGKMW) is proposed and proved to be a legal function in Kernel Method.The main purpose of this paper is to study clustering algorithm based on WGKMW. The purpose of this thesis is to enhance the properties and extend the applicability of SVC and then gives support to pattern analysis, Artificial Intelligence and Machine Learning. On the other hand, as an independent subject, kernel method is just at the initial stage and its capacity hasn't been fully discovered.The main researches that have been done are as follows:1. According to the defect of Gaussian kernel, a new kernel function called Weighted Gaussian Kernel with Multiple Widths is proposed in support vector clustering. The new kernel has more parameters which can be regulated to improve the generalization ability and learning capacity of kernel machine. 2. A clustering algorithm based on Weighted Gaussian Kernel with Multiple Widths is put forward by combining Weighted Gaussian Kernel with Multiple Widths with clustering analysis. Different widths reflect the difference between sample characteristics. According to nonlinear transformation and characteristics mentioned above, the separability of elements in feature space is enhanced when a set of data mapped into high-dimension feature space.3. It has effects on support vector clustering when kernel function parameters are changed. Experimental results show that WGKMW is better than Gaussian Kernel.