Enhancing in-tree-based clustering via distance ensemble and kernelization

Abstract

Abstract Recently, we have proposed a novel physically-inspired method, called the Nearest Descent (ND), which plays the role of organizing all the samples into an effective Graph, called the in-tree. Due to its effective characteristics, this in-tree proves very suitable for data clustering. Nevertheless, this in-tree-based clustering still has some non-trivial limitations in terms of robustness, capability, etc. In this study, we first propose a distance-ensemble-based framework for the in-tree-based clustering, which proves a very convenient way to overcome the robustness limitation in our previous in-tree-based clustering. To enhance the capability of the in-tree-based clustering in handling extremely linearly-inseparable clusters, we kernelize the proposed ensemble-based clustering via the so-called kernel trick. As a result, the improved in-tree-based clustering method achieves high robustness and accuracy on diverse challenging synthetic and real-world datasets, showing a certain degree of practical value.