Reducing SVR Support Vectors by Using Backward Deletion

摘要

Support Vector Regression (SVR) is one of the most famous sparse kernel machines which inherits many advantages of Support Vector Machines (SVM). However, since the number of support vectors grows rapidly with the increase of training samples, sparseness of the SVR is sometimes insufficient. In this paper, we propose two methods which reduce the SVR support vectors using backward deletion. Experiments show our method can dramatically reduce the number of support vectors without sacrificing the generalization performance.