Abstract: The success of kernel methods depends on the kernel,thus a choice of a kernel and proper setting of its parameters are of crucial importance．Learning a kernel from the data requires evaluation measures to assess the quality of the kernel．Recently,kernel target alignment (KTA),which measures the degree of agreement between a kernel and a learning task,has been widely used for kernel selection because of its effectiveness and efficiency．However,it is reported that KTA is only a sufficient condition to select a good kernel,but not a necessary condition．The reason is that KTA is not invariant under data translation in the feature space.This paper proposed a new measure for kernel selection named kernel distance target alignment (KDTA)．The measure not only overcomes the limitations of KTA but also possesses other properties like simplicity and efficiency．Comparative experiments indicate that the new measure is a good indication of the superiority of a kernel.

Key words: Kernel method,Kernel evaluation measure,Kernel distance