AutoML Regression Service for Data Analytics and Novel Data-centric Visualizations
While Machine learning (ML) use has become prevalent across most domains, there is a growing gap between programmers and non-programmers in their use of ML. Indeed, choosing the best models, applying the models, and verifying their quality is out of reach for individuals who rely on this kind of quantitative analysis but have limited programming experience–particularly those in the natural and social sciences. Automatic ML (AutoML) is supposed to be the stopgap giving non-programmers the ability to fully use ML, but in practice, these tools fall short. In response to this challenge, we built a data-centric machine learning web service we call ``AReS'' that both simplifies and streamlines the entire ML pipeline. AReS at its simplest only requires data. It chooses among dozens of diverse regression algorithms, picking the best. AReS gives both symbolic and visual assessments of the model’s performance through novel data-centric visualizations that provide insight into the data itself, both individual points and collections. To validate AReS, two cases using real-world Kaggle competitions (kaggle.com) are studied with AReS' default settings. AReS delivers competitive results in both but is among the best results in one. This paper’s novel web service, AReS, can be accessed at https://dalkilic.luddy.indiana.edu/.