K-ensembles: Combining K-means and ensemble learning for clustering machine learning models

Clustering is an unsupervised learning technique to group samples into clusters. Clusteringhas been applied to several data types, but not yet to machine learning (ML)models. The motivation of this paper is clustering ML models. For this purpose, onegeneral idea is to apply k-means. However, averaging ML models is not possible.Accordingly, this paper proposes a novel clustering algorithm, namely k-ensembles.The idea is to apply an ensemble learning to approximate the average of ML models.The proposed method represents ML models as rankings and computes their similarity.The other parts are the same as in k-means clustering. The experiment appliesK-ensembles for clustering one-class classification(OCC) models. The centroids of Kensemblesclustering can act as ensemble models. Moreover, K-ensembles can applyclustering on datasets by treating them as OCC models. Furthermore, K-ensembles canextract feature vectors from ML models using similarities to centroids.