A Closer Look at the Kernels Generated by the Decision and Regression Tree Ensembles

Tree ensembles can be interpreted as implicit kernel generators, where the ensuing proximity matrix represents data-driven tree ensemble kernel. Focus of our work is utility based generators that (in conjunction with a regularized linear model) enable learning. We elucidate performance random forest (RF) and gradient boosted (GBT) kernels in comprehensive simulation study comprising continuous binary targets. show for targets (regression), this learning approach competitive to respective higher dimensional scenarios, particularly cases larger number noisy features. For target (classification), their exhibit comparable performance. provide results from several real life datasets regression classification relevant biopharmaceutical biomedical applications, are line simulations how these insights may leveraged practice.We discuss general applicability extensions survival interpretable landmarking regression. Finally, we outline future research due feature space partitionings.