Identifying Red Variables in the Northern Sky Variability Survey

We present a catalog of 8678 slowly varying stars with near-infrared colors corresponding to the evolved asymptotic giant branch population. Objects were selected from the Northern Sky Variability Survey (NSVS) covering the entire sky above declination δ=-38^deg in a single unfiltered photometric band corresponding to a V-band magnitude range of ~8-15.5 mag. After quality cuts, the number of measurements for a typical star is approximately 150, but it ranges up to ~1000 for high-declination stars. We show that the use of support vector machines, a modern machine-learning algorithm, can reliably distinguish Mira variables from other types of red variables, namely, semiregular and irregular. We also identify a region of parameter space that is dominated by carbon stars. Our classification is based on period, amplitude, and three independent colors possible with photometry from the NSVS and the Two Micron All Sky Survey. The overall classification accuracy is ~90% despite the relatively short survey baseline of 1 yr and limited set of features. There are 6474 stars in our sample without identifications in the General Catalogue of Variable Stars, which, as such, are most likely new discoveries. Period-amplitude and period-color diagrams of both our previously known and newly identified Mira stars are in good agreement with published studies based on smaller samples.

Based on observations obtained with the ROTSE-I Robotic Telescope operated at Los Alamos National Laboratory.