Searching for Pulsating Stars Using Clustering Algorithms

Abstract

Clustering algorithms constitute a multi-disciplinary analytical tool commonly used to summarise large data sets. Astronomical classifications are based on similarity, where celestial objects are assigned to a specific class according to specific physical features. The aim of this project is to obtain relevant information from high-dimensional data (at least three input variables in a data-frame) derived from stellar light-curves using a number of clustering algorithms such as K-means and Expectation Maximisation. In addition to identifying the best performing algorithm, we also identify a subset of features that best define stellar groups. Three methodologies are applied to a sample of Kepler time series in the temperature range 6500–19,000 K. In that spectral range, at least four classes of variable stars are expected to be found: δ Scuti, γ Doradus, Slowly Pulsating B (SPB), and (the still equivocal) Maia stars.