Addressing the general problem of data clustering, we propose to group the elements of a data set with respect to their location within their minimum enclosing ball. In particular, we propose to cluster data according to their distance to the center of a kernel minimum enclosing ball. Focusing on kernel minimum enclosing balls which are computed in abstract feature spaces reveals latent structures within a data set and allows for applying our ideas to non-numeric data. Results obtained on image-, text-, and graph-data illustrate the behavior and practical utility of our approach.
