The Classification of Stellar Systems Through Singular Spectrum Analysis

• Activity Number: 254 - Contributed Poster Presentations: Section on Statistical Learning and Data Science
• Type: Contributed
• Date/Time: Monday, July 30, 2018 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Kevin Matheson* and Kevin Covey and Kimihiro Noguchi
• Companies: Western Washington University and Western Washington University and Western Washington University
• Keywords: SSA; machine learning; Singular Spectrum Analysis

Abstract

Through singular spectrum analysis (SSA) and an automated machine learning program called auto-sklearn, infrared spectrum data are analyzed to classify spectroscopic stellar systems as being binary or non-binary stars. By viewing the cross correlation function values of the infrared spectrum data at multiple visits of already classified stars as time series, eigenvalues of their trajectory matrices are calculated using SSA in the first step. Then, using auto-sklearn, common features in these eigenvalues among known binary stars and confident non-binary stars are sought to assign probabilities of binarity for unclassified stars. Through the use of this method, hundreds of previously unclassified stars from the Sloan Digital Sky Survey are identified as potentially binary stars.