Citizen ASAS-SN Data Release I: Variable Star Classification Using Citizen Science

TL;DR

This paper introduces Citizen ASAS-SN, a citizen science project that classifies variable stars using ASAS-SN data, achieving high accuracy and discovering thousands of new variables, while also enhancing machine learning classifiers.

Contribution

The study demonstrates the effectiveness of citizen science in classifying variable stars and improves machine learning models using volunteer classifications and 'Junk' data.

Findings

Volunteers classified over 10,000 new variable stars with high accuracy.

User activity correlates with classification accuracy and agreement.

Machine learning classifiers are improved using 'Junk' classifications from volunteers.

Abstract

We present the first results from Citizen ASAS-SN, a citizen science project for the All-Sky Automated Survey for Supernovae (ASAS-SN) hosted on the Zooniverse platform. Citizen ASAS-SN utilizes the newer, deeper, higher cadence ASAS-SN $g$-band data and tasks volunteers to classify periodic variable star candidates based on their phased light curves. We started from 40,640 new variable candidates from an input list of ${\sim} 7.4$ million stars with $\delta < -60^\circ$ and the volunteers identified 10,420 new discoveries which they classified as 4,234 pulsating variables, 3,132 rotational variables, 2,923 eclipsing binaries, and 131 variables flagged as Unknown. They classified known variable stars with an accuracy of 89% for pulsating variables, 81% for eclipsing binaries, and 49% for rotational variables. We examine user performance, agreement between users, and compare the citizen science classifications with our machine learning classifier updated for the $g$-band light curves. In general, user activity correlates with higher classification accuracy and higher user agreement. We used the user's "Junk" classifications to develop an effective machine learning classifier to separate real from false variables, and there is a clear path for using this "Junk" training set to significantly improve our primary machine learning classifier. We also illustrate the value of Citizen ASAS-SN for identifying unusual variables with several examples.