Semi-supervised Least-squares Support Vector Regression Machines ★
Shuo Xu, Xin An, Xiaodong Qiao, Lijun Zhu, Lin Li · 2011
AbstractIn many real-world applications, unlabeled examples are inexpensive and easy to obtain. Semi-supervisedapproaches try to utilize such examples to boost the predictive performance. But previous researchmainly focuses on classification problem, and semi-supervised regression remains largely under-studied.In this work, a novel semi-supervised regression method, semi-supervised LS-SVR (S 2 LS-SVR), isproposed on the basis of LS-SVR. Similar to the LS-SVR, one only solves a convex linear system in thetraining phrase too, thus largely speeding up training. Experimental results on corn data set indicatethat our approach is feasible and efficient. Keywords : Semi-supervised Learning; Regression Problem; Least-squares Support Vector RegressionMachine (LS-SVR); Semi-supervised LS-SVR (S 2 LS-SVR) 1 Introduction Traditionally, hypotheses are learned from a large number of training examples, in each of whicha label is attached. For classification problem, the label indicates the category into which thecorresponding example falls; for regression problem, the label is a real-value. Most machinelearning methods rely on the availability of large labeled examples, since the larger the number