Meta-Learning Based Feature Selection for Clustering

Oleg Taratukhin, Sergey Muravyov

Abstract: Clustering is a highly demanded task nowadays. However, it requires the attention of human experts and it might certainly benefit from automation. This paper presents a method to perform simultaneous automatic algorithm selection and hyperparameter tuning with feature selection for clustering. The algorithm also features a meta-model to predict promising algorithm configurations based on dataset properties. Experimental results are provided for a set of benchmark datasets, it is shown that the proposed method outperforms known alternatives.

External IDs:doi:10.1007/978-3-030-91608-4_54