摘要：Taking motivation from ɛ-insensitive twin supportvector regression (ɛ-TSVR) and the projection idea, this paperproposes a novel ɛ-twin projection support vector regressionmodels, called ɛ-TPSVR. The proposed ɛ-TPSVR, which isbased on ɛ-TSVR, determines the regression function through apair of nonparallel hyperplanes solved by two smaller sizedquadratic programming problems. Different from ɛ-TSVR, aprojection axis is sought for each optimization problem ofɛ-TPSVR such that the variance of the projected points isminimized. Therefore, the empirical correlation coefficientbetween each hyperplane and the projected inputs can beoptimized. The experimental results indicate that the proposedɛ-TPSVR obtains the better prediction performance than TSVRand ɛ-TSVR methods that were widely adopted.
