Presentation #541.06 in the session “Computational Augmentation to Surveys and Science Programs”.
Due to rapidly increasing data volume in astronomy, machine learning techniques are necessary for the classification of large samples of different types of objects. Evolved stars are the major dust producers in the Universe and they have an important role in the enrichment of the interstellar medium. The automated classification of mid-IR photometry and spectra is therefore essential for the identification of dusty evolved stars in nearby galaxies. As a first step, we test the Support Vector Machine classification techniques on a well-studied sample of evolved stars from the SAGE-SMC (Surveying the Agents of Galaxy Evolution; Meixner et al. 2006, Gordon et al. 2010, Boyer et al. 2011) sample. We determine linear boundaries to separate various types of evolved stars in near-IR and mid-IR colour-magnitude diagrams (CMDs) using 2MASS JHKs photometry along with the Spitzer Space Telescope IRAC (3.6, 4.5, 5.8, and 8.0 micron) and MIPS 24 micron band photometry. We restrict our sample to sources brighter than 13.0 mag in Ks band, reducing contamination from red giant branch stars. We perform supervised classification on the resulting data set of 11596 stars, the Support Vector Machine (SVM) technique using a linear kernel. We also repeat our analysis using unsupervised classification schemes. Our results agree very well with the CMD-based classifications of Boyer et al. (2011). We obtained an overall accuracy of 98% with the SVM method. The lowest accuracy (94%) is obtained for the extremely dusty AGB candidates. As a next step, we will incorporate spectroscopic data from the SAGE-Spec program (Kemper et al. 2010, Woods et al. 2011) as well as variability information to enhance our classification accuracy. This work will allow us to identify interesting targets in nearby galaxies for detailed follow-up with upcoming facilities such as the James Webb Space Telescope.