Classification of Variable Stars using Convolutional Neural Network

Author

Abhina Premachandran Bindu, CUNY City CollegeFollow

Date of Award

2025

Document Type

Thesis

Department

Computer Science

First Advisor

Greg Olmschenk

Second Advisor

Zhigang Zhu

Keywords

Machine learning, Convolutional Neural Networks, Variable Stars, Transiting Exoplanet Survey Satellite, Variable Star Catalog

Abstract

This research focuses on developing Convolutional Neural Networks (CNNs), for the process of classifying and identifying variable stars through the analysis of unprocessed light curves from Transiting Exoplanet Survey Satellite (TESS). As astronomical data is becoming increasingly complex, and as advanced missions deploy sophisticated instruments for data collection, both the quality and quantity of the data are improving at a rapid pace. This has created an urgent need to automate the analysis process using efficient and effective methods, such as those based on machine learning. While previous research has explored machine learning approaches, there has been limited focus on implementing CNNs for this task. Among the limited studies employing CNNs, most either relied on preprocessing techniques that introduced artificial associations within the data, or used data beyond raw light curves or used a CNN structure that wasn't deep enough to study the complex patterns in light curves. The data used in this work is from TESS, an MIT-led NASA mission dedicated to observing millions of stars to detect transiting exoplanets. With photometrically precise, high-cadence brightness data, this rich dataset is ideal for detecting exoplanets and well-suited for studying variable stars, as it captures subtle brightness changes. However, using TESS light curves for classifying variable stars remains underexplored, in large part due to the scale of the dataset. Our classifier is significantly fast, and on a single GPU, it takes on the order of only a few tens of milliseconds to process a single light curve. Therefore, our work focuses on developing an efficient and reliable tool to automate the classification of TESS light curves as variable stars and to create a catalog of newly found variable star candidates. Our CNN classifier model can classify and identify four different types of variable stars: Delta Scuti variables, RR Lyrae variables, rotation modulations, and eclipsing binaries from a sample among negative or light curves whose variability is unknown. The current model has identified 2,569 new variable star candidates from a total of 209,658 light curves.

Recommended Citation

Premachandran Bindu, Abhina, "Classification of Variable Stars using Convolutional Neural Network" (2025). CUNY Academic Works.
https://academicworks.cuny.edu/cc_etds_theses/1268