Abstract: This paper presented a new set of kernel function-Legendre kernel function based on Legendre polynomial.The performance and robustness of the presented kernel were investigated on bi-spiral benchmark data set as well as five data sets from the UCI benchmark repository. I}he experiment results demonstrate that the presented kernel has competitive robust and generalization performance compared with commonly used kernel functions (polynomial kernel and Radial Basis Function ctc.).Moreover, the I_egendre kernel has one parameter which is only chosen from natural number, thus parameter optimization is facilitated greatly.

Key words: Support vector machine, Kernel function, Model selection