A local minimum sampling strategy for the construction of boundaries using support vector machines

This work presents a study of a new adaptive sampling strategy for the construction of explicit boundaries using Support Vector Machines (SVMs), referred to as Explicit Design Space Decomposition (EDSD). The new adaptive sampling strategy called local minimum point is based on choosing the sample on the current SVM boundary with the minimum distance to the most important support for the current SVM. The proposed scheme can not only prevent locking the SVM boundary and balance the negative class and positive one, but also be used as a main refining scheme. And the ability of this scheme was illustrated by two analytical functions, a three dimensional problem and a five dimensional problem.