ON MACHINE-LEARNED CLASSIFICATION OF VARIABLE STARS WITH SPARSE AND NOISY TIME-SERIES DATA

Abstract

With the coming data deluge from synoptic surveys, there is a growing need\nfor frameworks that can quickly and automatically produce calibrated\nclassification probabilities for newly-observed variables based on a small\nnumber of time-series measurements. In this paper, we introduce a methodology\nfor variable-star classification, drawing from modern machine-learning\ntechniques. We describe how to homogenize the information gleaned from light\ncurves by selection and computation of real-numbered metrics ("feature"),\ndetail methods to robustly estimate periodic light-curve features, introduce\ntree-ensemble methods for accurate variable star classification, and show how\nto rigorously evaluate the classification results using cross validation. On a\n25-class data set of 1542 well-studied variable stars, we achieve a 22.8%\noverall classification error using the random forest classifier; this\nrepresents a 24% improvement over the best previous classifier on these data.\nThis methodology is effective for identifying samples of specific science\nclasses: for pulsational variables used in Milky Way tomography we obtain a\ndiscovery efficiency of 98.2% and for eclipsing systems we find an efficiency\nof 99.1%, both at 95% purity. We show that the random forest (RF) classifier is\nsuperior to other machine-learned methods in terms of accuracy, speed, and\nrelative immunity to features with no useful class information; the RF\nclassifier can also be used to estimate the importance of each feature in\nclassification. Additionally, we present the first astronomical use of\nhierarchical classification methods to incorporate a known class taxonomy in\nthe classifier, which further reduces the catastrophic error rate to 7.8%.\nExcluding low-amplitude sources, our overall error rate improves to 14%, with a\ncatastrophic error rate of 3.5%.\n