Research of Support Vector Regression Algorithm Based on Granularity

Abstract

—A regression method of Support Vector Machines in the case of a large number of sample data. Hierarchies of various granularities for the data set are constructed by density clustering algorithm. In coarse-granularity level, abnormal sample data are excluded, while part of dense repeated samples are removed in fine-granularity level. After pretreating the sample set by the method mentioned above, Support Vector Regression is trained to construct a regression model. In this paper, the prediction model of coke mechanical strength is established by the means. The result indicates that Support Vector Regression Algorithm based on granularity has low computational complexity and high speed, moreover eliminating noise sample data and removing the dense samples do not affect the distribution and prediction effect of the original sample set. It is an effective measure of regression with the large sample data.