Kernel Parameter Selection of Gaussian Process Classification Model Based on DBTC Criterion

This paper proposes a new kernel parameter selection method of GPC model based on DBTC criterion.After the kernel parameter is used as independent variable, and the DBTC is used as induced variable, we obtain object function that DBTC is varied with kernel parameter.Following that, conjugate gradient method is utilized to calculate the exterma of object function.Finally, the optimal value of kernel parameter is obtained.Experiments illustrates that the proposed method achieved comparable classification accuracy to traditional method.However, the time consuming in parameter section is sharply shortened.Consequently, the training speed of GPC model is improved.