Metallicity Effects on Machine Learning Classification of Dusty Stellar Sources in the Magellanic Clouds

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Ghaziasgar, Sepideh, Abdollahi, Mahdi, Javadi, Atefeh, van Loon, Jacco Th., McDonald, Iain, Oliveira, Joana, Khosroshahi, Habib G.
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909905273749504
author Ghaziasgar, Sepideh
Abdollahi, Mahdi
Javadi, Atefeh
van Loon, Jacco Th.
McDonald, Iain
Oliveira, Joana
Khosroshahi, Habib G.
author_facet Ghaziasgar, Sepideh
Abdollahi, Mahdi
Javadi, Atefeh
van Loon, Jacco Th.
McDonald, Iain
Oliveira, Joana
Khosroshahi, Habib G.
contents Differences in metallicity between the Large Magellanic Cloud (LMC) and the Small Magellanic Cloud (SMC) offer an opportunity to examine whether environmental metallicity affects the performance of machine learning models in classifying dusty stellar sources. The five stellar classes studied include young stellar objects (YSOs), red supergiants (RSGs), post-asymptotic giant branch stars (PAGBs), and oxygen- and carbon-rich asymptotic giant branch stars (OAGBs and CAGBs), which are key phases of stellar evolution involved in dust production. Using spectroscopically labeled data from the Surveying the Agents of Galaxy Evolution (SAGE) project, we trained and evaluated a probabilistic random forest (PRF) classifier with four approaches: (1) separate training on LMC and SMC, including all five classes, (2) excluding the underpopulated PAGB class, (3) combined LMC and SMC datasets, and (4) cross-galaxy training and testing. The model achieved 93\% accuracy on the SMC and 88\% on the LMC across all five classes. In the SMC, PAGB sources were misclassified as YSOs, mainly because of their small sample size (4 objects). When PAGB was excluded, both the LMC and the SMC reached 92\% accuracy. A combined dataset produced the same accuracy, and cross-galaxy training yielded similar results, indicating that metallicity does not significantly impact model performance. A comparison of absolute CMDs for the LMC and SMC confirms their similarity in stellar populations. These findings suggest that environmental metallicity has little effect on ML-based classification of dusty stellar sources, supporting the use of combined datasets and cross-galaxy models in low-metallicity environments.
format Preprint
id arxiv_https___arxiv_org_abs_2511_12210
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Metallicity Effects on Machine Learning Classification of Dusty Stellar Sources in the Magellanic Clouds
Ghaziasgar, Sepideh
Abdollahi, Mahdi
Javadi, Atefeh
van Loon, Jacco Th.
McDonald, Iain
Oliveira, Joana
Khosroshahi, Habib G.
Astrophysics of Galaxies
Differences in metallicity between the Large Magellanic Cloud (LMC) and the Small Magellanic Cloud (SMC) offer an opportunity to examine whether environmental metallicity affects the performance of machine learning models in classifying dusty stellar sources. The five stellar classes studied include young stellar objects (YSOs), red supergiants (RSGs), post-asymptotic giant branch stars (PAGBs), and oxygen- and carbon-rich asymptotic giant branch stars (OAGBs and CAGBs), which are key phases of stellar evolution involved in dust production. Using spectroscopically labeled data from the Surveying the Agents of Galaxy Evolution (SAGE) project, we trained and evaluated a probabilistic random forest (PRF) classifier with four approaches: (1) separate training on LMC and SMC, including all five classes, (2) excluding the underpopulated PAGB class, (3) combined LMC and SMC datasets, and (4) cross-galaxy training and testing. The model achieved 93\% accuracy on the SMC and 88\% on the LMC across all five classes. In the SMC, PAGB sources were misclassified as YSOs, mainly because of their small sample size (4 objects). When PAGB was excluded, both the LMC and the SMC reached 92\% accuracy. A combined dataset produced the same accuracy, and cross-galaxy training yielded similar results, indicating that metallicity does not significantly impact model performance. A comparison of absolute CMDs for the LMC and SMC confirms their similarity in stellar populations. These findings suggest that environmental metallicity has little effect on ML-based classification of dusty stellar sources, supporting the use of combined datasets and cross-galaxy models in low-metallicity environments.
title Metallicity Effects on Machine Learning Classification of Dusty Stellar Sources in the Magellanic Clouds
topic Astrophysics of Galaxies
url https://arxiv.org/abs/2511.12210