Comparison of Photometric and Spectroscopic Labels in Classifying Dusty Stellar Sources Using Machine Learning in the Magellanic Clouds
Sepideh Ghaziasgar, Mahdi Abdollahi, Atefeh Javadi, Jacco Th. van Loon, Iain McDonald, Joana Oliveira, Habib G. Khosroshahi
公開日: 2025/9/5
Abstract
Dusty stellar sources, including young stellar objects (YSOs) and evolved stars such as oxygen- and carbon-rich AGBs (OAGBs, CAGBs), red supergiants (RSGs), and post-AGB stars (PAGBs), play a key role in the chemical enrichment of galaxies. Photometric surveys in the Magellanic Clouds have cataloged many such objects, but their classifications are often uncertain due to overlaps between populations. We trained machine learning models on spectroscopically labeled data from the SAGE project and applied them to photometric catalogs. The spectroscopic model achieves about 89\% accuracy. Applied to photometric labels, nearly all OAGBs are correctly identified, and YSOs have a 95\% confirmation rate. In contrast, 16\% of CAGBs are reclassified as OAGBs, only 8\% of RSGs retain their labels, and fewer than half of PAGBs are confirmed. Photometry is thus reliable for abundant populations with distinct signatures, but spectroscopic confirmation remains essential for rare or overlapping stellar classes.