Fuzzy rough based regularization in Generalized Multiple Kernel Learning

Generalized Multiple Kernel Learning (GMKL) has been proposed in the literature for feature selection. GMKL learns linear, product and exponential combinations of given base kernels which makes it more robust and efficient than traditional Multiple Kernel Learning (MKL). GMKL has been shown to be a good tool for feature selection as well. Time taken for the convergence of GMKL depends upon the initialization of kernel weights. Optimization schemes in GMKL initialize kernel weights randomly. This produces variability in convergence time.In this paper, we propose a Fuzzy Rough Set based kernel weight initialization for GMKL (FR-GMKL). We show that this results in faster convergence than that obtained by random initialization in GMKL while retaining same level of accuracy. We also show that the computation time of our proposed method is lower than that obtained through Quadratic Programming Feature Selection (QPFS) based as well as Maximal Relevance (MaxRel) based initialization of the regularization parameter. The performance of our proposed method is evaluated using five benchmark binary classification datasets and three benchmark multi-class classification datasets from the UCI repository.