Automatic selection of clustering algorithms using supervised graph embedding

The widespread adoption of machine learning (ML) techniques and the extensive expertise required to apply them have led increased interest in automated ML solutions that reduce need for human intervention. One main challenges applying previously unseen problems is algorithm selection – identification high-performing algorithm(s) a given dataset, task, evaluation measure. This study addresses challenge data clustering, fundamental task mining aimed at grouping similar objects. We present MARCO-GE, novel meta-learning approach recommendation clustering algorithms. MARCO-GE first transforms datasets into graphs then utilizes graph convolutional neural network technique extract their latent representation. Using embedding representations obtained, trains ranking meta-model capable accurately recommending top-performing algorithms new dataset An on 210 datasets, 17 algorithms, 10 measures demonstrates effectiveness our its superiority terms predictive generalization performance over state-of-the-art approaches.