Unsupervised AutoML: a study on automated machine learning in the context of clustering

Abstract

The modelling of end-to-end Machine Learning processes and methods is not only computationally intensive, but also requires expertise in Data Science and often domain knowledge of the problem. To overcome this adversity, a relatively new field of research has emerged called Automated Machine Learning (AutoML). The main focus of the domain is to discover an automated way to build Machine Learning pipelines given a Machine Learning task and an input data set. While all AutoML systems currently focus on the task of supervised learning, unsupervised learning remains an unexplored and unsolved problem. This thesis aims to provide solutions for automating Machine Learning specifically for the case of unsupervised learning (clustering), in a domain-agnostic manner. This is achieved through a combination of state-of-the-art processes based on Meta Learning for Algorithm Selection and Bayesian Optimization for hyperparameter tuning. Experimentation results on real life datasets provide enough evidence that clustering is a process that can be fully automated.