Supervised Ensemble Classification of Kepler Variable Stars

TL;DR

This paper introduces an ensemble machine learning approach to automatically classify approximately 150,000 Kepler variable stars into 14 categories, improving classification accuracy over previous methods.

Contribution

It expands existing methods by combining multiple classification techniques for better accuracy in large-scale stellar variability classification.

Findings

Achieved improved classification rates with ensemble approach.

Successfully classified 150,000 stars into 14 variable classes.

Demonstrated effectiveness of combined techniques in large datasets.

Abstract

Variable star analysis and classification is an important task in the understanding of stellar features and processes. While historically classifications have been done manually by highly skilled experts, the recent and rapid expansion in the quantity and quality of data has demanded new techniques, most notably automatic classification through supervised machine learning. We present an expansion of existing work on the field by analyzing variable stars in the {\em Kepler} field using an ensemble approach, combining multiple characterization and classification techniques to produce improved classification rates. Classifications for each of the roughly 150,000 stars observed by {\em Kepler} are produced separating the stars into one of 14 variable star classes.