Evolutionary Optimisation of Kernel and Hyper-Parameters for SVM

Support Vector Machines (SVMs) concern a new generation learning systems based on recent advances in statistical learning theory. A key problem of these methods is how to choose an optimal kernel and how to optimise its parameters. A (multiple) kernel adapted to the problem to be solved could improve the SVM performance. Therefore, our goal is to develop a model able to automatically generate a complex kernel combination (linear or non-linear, weighted or un-weighted, according to the data) and to optimise both the kernel parameters and SVM parameters by evolutionary means in a unified framework. Furthermore we try to analyse the architecture of such kernel of kernels (KoK). Numerical experiments show that the SVM algorithm, involving the evolutionary KoK performs statistically better than some well-known classic kernels and its architecture is adapted to each problem.