Fast tuning of SVM kernel parameter using distance between two classes

In the construction of support vector machines (SVM) an important step is to select the optimal kernel parameters. This letter proposes using the distance between two classes (DBTC) in the feature space to help choose kernel parameters. Based on the proposed method, the DBTC function is approximated accurately with sigmoid function. The computation complexity decreases significantly since training SVM and the test with all parameters are avoided. Empirical comparisons demonstrate that the proposed method can choose the parameters precisely, and the computation time decreases dramatically.