The Research Of Fusion Algorithms For Support Vector Machine And K-means Clustering

In this paper, we research fusion algorithms of support vec-tor machine and K-means clustering algorithm, on the basis of thefoundational theories of support vector machine and K-means clus-tering algorithm, introduce the theoretical knowledge of improvingalgorithms carefully, and verify the e?ectiveness of the algorithms byexperiment.Support Vector Machine(SVM) is a new and very e?ective methodof machine learning developed on Statistical Learning Theory, whichconcludes optimization, kernel and the ability of the best promotingand so on. It solves many practical problems which troubled manylearning methods in the past, such as small sample, non-linear, over-study, high-dimension and local minimum point. Compared with theother machine learning methods, the support vector machine has in-comparable advantages in many aspects, but it also has its own lim-itations. For the sensitive problem to noises and outliers, we haveproposed a approach of removing noises and outliers for SVM basedon fuzzy membership which adds fuzzy membership in one-class SVMunder linear programming, and verify the e?ectiveness of the algorithmin a variety of data sets.K-means clustering algorithm is a simple, fast and classical algo-rithm to solve the problem of clustering. If the samples are intensive and linear among the categories, it is the best method, but if they arenon-linear, the clustering e?ect isn't perfect. Against this problem,we have proposed K-means clustering algorithm based on SVM, whichadds one-class SVM in K-means clustering algorithm. In this paper, weconducts experiments respectively on a synthetic data set(Delta Set)and a UCI data set(Iris Data), they prove that its clustering accuracyimproved obviously compared with other algorithms. Compared withSVM under quadratic programming, SVM under linear programmingdoesn't only improve the accuracy of clustering, but also reduce thecomplexity of the algorithm greatly.