Clustering using a Sum-Over-Forests weighted kernel k-means approach

Sommer, Felix [UCL] Fouss, François [UCL] Saerens, Marco [UCL]

We develop a weighted kernel k-means approach for clustering using the Sum-Over-Forests density index as node weights. Furthermore, we implement and test the developed algorithm and vary the used kernels to ascertain the algorithms functionality. The algorithm gives good results, but is strongly dependent on the kernel parameters as well as the density index parameter, both of which have to be tuned for optimal results. We achieve a reduction in algorithm runtime, albeit at the price of statistical measure scores.