[View Abstracts by Session]

Session: Stars, Planets and the Interstellar Medium

Name: Dr. Grigoris Maravelias (National Observatory of Athens & FORTH)
Coauthors: Bonanos Alceste (National Observatory of Athens)
de Wit Stephan (National Observatory of Athens, National & Kapodistrian Univ. of Athens)
Tramper Frank (KU Leuven)
Munoz-Sanchez Gonzalo (National Observatory of Athens, National & Kapodistrian Univ. of Athens)
Zapartas Manos (National Observatory of Athens)
Christodoulou Evangelia (National Observatory of Athens, National & Kapodistrian Univ. of Athens)
Antoniadis Konstantinos (National Observatory of Athens, National & Kapodistrian Univ. of Athens)
Bonfini Paolo (Ballista Technologies)
Yang Ming (National Astronomical Observatories, Chinese Academy of Sciences)
Type: Oral
Title: Investigating the populations of dusty evolved stars in various metallicities with machine learning
Abstract:

Although our knowledge on stellar evolution has improved dramatically over the last decades, both regarding single and binary evolutionary models, we still lack pieces of the puzzle. Mass loss is a key property to understand and characterize stellar evolution in particular beyond the main-sequence, as it determines the immediate environment around the source as well as its final fate. Unfortunately, there is a mismatch between the observational and theoretical values (derived from models). Even worse, the episodic mass loss in evolved massive stars, although definitely present observationaly, is not included in the models. Therefore, its importance of its role is currently undetermined. A major hindrance is the lack of large samples of classified stars. In the framework of the ASSESS project, we attempted to address this by applying machine-learning techniques to IR (Spitzer) and optical (Pan-STARRS) photometry, coupled with Gaia astrometry to detect foreground sources. Using color indices as features we utilized an ensemble approach (combining the probabilities from three different algortihms, Support Vector Machines, Random Forest, and Multilayer Perceptron). The three supervised algorithms were trained on M31 and M33 sources with known spectral classification, and we grouped the sources into Blue/Yellow/Red Supergiants, Luminous Blue Variables, classical Wolf-Rayet stars, B[e] Supergiants, and a class for outliers (e.g. background galaxies, AGNs). We then applied the ensemble classifier to about one million Spitzer point sources from 25 galaxies, spanning a range of 1/15 to ~3 Zsolar. Only a tiny fraction (~0.5%) of these sources have spectral classification. We delivered relatively reliable classifications for ~30% of the Spitzer point sources, and, therefore, providing the most numerous catalog of massive evolved stars compiled for these galaxies to-date. Equipped with spectral classifications we investigated the occurrence of these classes and ratios with respect to metallicity environments.