A multi-agent system for automated machine learning

Abstract

Machine Learning (ML) focuses on giving machines the ability to forecast, predict, or classify without being explicitly programmed to do so. To achieve such goals, large amounts of data are used to conceive models that can adapt to unseen data and to new scenarios. However, applying ML models to real-world business domains is a resource-intensive and time-consuming effort. Automated machine learning (AutoML) emerged as a way to ease such processes. With this in mind, this study introduces a multi-agent system (MAS) that autonomously go through the entire ML pipeline, with different entities being responsible for the data collection process, for pre-processing the data, and for deploying the best candidate ML model. The conceived MAS is currently implemented in a real-world setting, addressing important societal challenges raised by big urban centers. The obtained results show that this solution proved to be beneficial not only for the data collection and pre-processing tasks, but also for the automated execution of ML models.