Kernel-Based Clustering Driven by Density Index

We propose a clustering method that can deal with non-linearly separable groups by means of a search process in which we look for the best values for parameters of a set of kernel functions that induce, on the objects to be clustered, the best partition relative to quality in terms of a density-based criterion. In summary, our proposal is an iterative adaptation of a set of kernel parameters, guided by a density-based criterion, which is able to yield clustering solutions that improve those results obtained from the separate application of kernel and density-based methods from the state-of-the-art.