An Automated Machine Learning Platform for Non-experts

Abstract

With successful applications of machine learning to various domains, there have been large demands on developing machine learning-based applications. Automated machine learning is crucial to meet the demand because there are not sufficiently many expert machine learning developers to support such various demands. This paper presents an automated machine learning platform which gets some basic information about a task from nonexpert developer and examines several candidate models to develop an effective machine learning model. To choose some candidate machine learning pipeline for the given task, the platform makes use of the HTN-based plans to describe the machine learning plans along with its application conditions. The prototype system has been developed to mainly support machine learning models for tabular data including time-series data.