Machine learning based stellar classification with highly sparse photometry data.

Area of Science:

• Astronomy
• Astrophysics
• Machine Learning

Background:

• Accurate stellar classification is essential for studying stellar evolution.
• Large-scale astronomical surveys necessitate automated classification methods.
• Current methods face challenges with vast datasets and missing information.

Purpose of the Study:

• To develop and test a Machine Learning (ML) model for classifying stars into nine distinct classes using photometric data.
• To evaluate the impact of data sparsity and class imbalance on ML model performance.
• To explore the utility of various data features, including photometric measurements and Galactic position.

Main Methods:

• Utilized a multi-class, multi-label XGBoost (Extreme Gradient Boosting) Machine Learning algorithm.
• Employed the PySSED spectral-energy-distribution fitting algorithm for data analysis.
• Trained the classifier on subsets of the SIMBAD astronomical database, addressing data sparsity and class imbalance.

Main Results:

• The ML classifier achieved an accuracy of approximately 0.7 and a macro F1 score of 0.61.
• Performance varied based on the inclusion or exclusion of specific variables.
• Increasing sample size for a star type significantly improved model performance for that type.

Conclusions:

• This research presents a proof of feasibility for using ML to classify stars based on photometric data.
• The current model's accuracy is insufficient for reliable, real-world stellar classification.
• Further development is needed to enhance the accuracy and robustness of ML-based stellar classification systems.