Evolution of kernels for support vector machine classification on large datasets

Abstract

Kernel selection is a main factor in the designing of support vector machines. Evolutionary techniques have been applied to select the fittest kernel for specific classification problems. However, technical issues emerge when attempting to apply this methodology to deal with large datasets. On the other hand, a new method for improving the training time of support vector machines was recently developed. In this chapter, the new method is integrated in a kernel evolution scheme. Ten benchmark datasets are tested. Results indicate that the new method speeds up the evolution process when datasets are greater than 1000 instances.