Intelligent combination of kernels information for improved classification

In this paper, we are proposing a combination scheme of kernels information of support vector machines (SVMs) for improved classification task using genetic programming. In the scheme, first, the predicted information is extracted by SVM through the learning of different kernel functions. GP is then used to develop an optimal composite classifier (OCC) having better performance than individual SVM classifiers. The experimental results demonstrate that OCC is more effective, generalized and robust. Specifically, it attains high margin of improvement at small features. Another side advantage of our GP based intelligent combination scheme is that it automatically incorporates the issues of optimal kernel and model selection to achieve a higher performance prediction model.