Automatic SVM Kernel Function Construction Based on Gene Expression Programming

Traditional support vector machine needs pre-assumed kernel functions. This paper proposes a method via gene expression programming to automatically construct the kernel. The contributions of this paper include: (1) proposing the concepts of GEP kernel and kernel tree; (2) proposing the properties of GEP kernel and the kernel relation theorem; (3) proposing GEP based support vector machine (KGEP-SVM), (4) decoding kernel individual algorithm (DKIA) and kernel operators operating algorithm (KOOA), and (5) extensive experiments show that the average accuracy of the method is increased by 4% and generation of GEP kernel is about 150.