The ASAS-SN Catalog of Variable Stars II: Uniform Classification of 412,000 Known Variables

TL;DR

This paper presents a homogeneous classification of approximately 412,000 variable stars from the VSX catalog using the ASAS-SN survey data and an advanced machine learning classifier, improving the understanding of variable star properties.

Contribution

The study introduces a uniform classification method for a large set of variable stars, deriving periods, reclassifying many sources, and providing comprehensive data accessible through an online database.

Findings

Achieved an F1 score of 99.4% with the classifier.

Derived periods for about 52,000 variables lacking them.

Reclassified approximately 17,000 sources into new groups.

Abstract

The variable stars in the VSX catalog are derived from a multitude of inhomogeneous data sources and classification tools. This inhomogeneity complicates our understanding of variable star types, statistics, and properties, and it directly affects attempts to build training sets for current (and next) generation all-sky, time-domain surveys. We homogeneously analyze the ASAS-SN V-band light curves of ${\sim}412,000$ variables from the VSX catalog. The variables are classified using an updated random forest classifier with an $F_1$ score of 99.4\% and refinement criteria for individual classifications. We have derived periods for ${\sim}52,000$ variables in the VSX catalog that lack a period, and have reclassified ${\sim} 17,000$ sources into new broad variability groups with high confidence. We have also reclassified ${\sim} 94,000$ known variables with miscellaneous/generic classifications. The light curves, classifications, and a range of properties of the variables are all available through the ASAS-SN variable stars database (https://asas-sn.osu.edu/variables). We also include the V-band light curves for a set of ${\sim}4,000$ rare variables and transient sources, including cataclysmic variables, symbiotic binaries and flare stars.