Meta-learning Based Evolutionary Clustering Algorithm

Dmitry Tomp, Sergey Muravyov, Andrey Filchenkov, Vladimir Parfenov

Abstract: In this work, we address the hard clustering problem. We present a new clustering algorithm based on evolutionary computation searching a best partition with respect to a given quality measure. We present 32 partition transformation that are used as mutation operators. The algorithm is a $$(1+1)$$ evolutionary strategy that selects a random mutation on each step from a subset of preselected mutation operators. Such selection is performed with a classifier trained to predict usefulness of each mutation for a given dataset. Comparison with state-of-the-art approach for automated clustering algorithm and hyperparameter selection shows the superiority of the proposed algorithm.