A Challenge of Developing a Classifier for Multi-Band Classification of Variable Stars

Xiao-Hui Xu, Qing-Feng Zhu, Xu-Zhi Li, Hang Zheng, Jin-Sheng Qiu · 2025

Read A Challenge of Developing a Classifier for Multi-Band Classification of Variable Stars by Xiao-Hui Xu, Qing-Feng Zhu, Xu-Zhi Li, Hang Zheng, Jin-Sheng Qiu (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-18731.txt.

Introduction

Xiao-Hui Xu and four co-authors explore whether a star classifier trained in one color of light can work in another, in a 2025 paper. Variable stars, whose brightness rises and falls, teach astronomers much about the galaxy, and sky surveys now harvest them by the millions. Sorting them by type takes classifiers that are costly to build for each survey's own band of light.

The team trained a random forest on brightness records from two surveys, in two different bands, then let it judge samples from a third survey's band. They compared feeding it raw magnitudes against normalized fluxes, and periods measured in one band against periods blended from several. The study maps what carries across bands and what breaks, a practical guide for astronomers who want one good classifier to serve many telescopes.

Catalog record

CREATORXiao-Hui Xu, Qing-Feng Zhu, Xu-Zhi Li, Hang Zheng, Jin-Sheng Qiu
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE52 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-18731

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A Challenge of Developing a Classifier for Multi-Band Classification of Variable Stars Xiao-Hui Xu, Qing-Feng Zhu, Xu-Zhi Li, Hang Zheng, Jin-Sheng Qiu Abstract Variable stars play a very important role in our understanding of the Milky Way and the universe. In recent years, many survey projects have generated a large amount of photometric data, necessitating classifiers that can quickly identify various types of variable stars. However, obtaining these classifiers often requires substantial manpower and computational resources. To conserve these resources, it would be best to have a classifier that can be used across surveys. We explore the possibility that a classifier created in one optical band can also work in other bands, likely from different survey facilities. We construct a random forest classifier based on photometric data in ASAS-SN V-band and OGLE I-band, and apply the classifier on ASAS-SN V-band and ZTF r-band light curves of variable star samples. We explore the classification differences of using the magnitude light-curves or the normalized flux light-curves, the periods derived from single band light-curves or the periods derived from the multi-band combined light-curves, and with or without color-related features. We find it feasible to develop a classifier capable of working in both V and r bands for certain types of variable stars, such as RRAB variables. For other types of variable stars, like Cepheids, the classifier is unable to make accurate identifications. \volnopage Vol.0 (20xx) No.0, 000–000 keywords methods: statistical — methods: data analysis — stars: variables: general 1 Introduction Variable stars are a large and important branch of time-varying sources that help us understand the Milky Way and the universe. Classical Cepheids (DCEPs) are the extra-galactic distance ladder and eventually lead to the determination of the Hubble constant (H 0 ) thanks to the well-defined period-luminosity and period-Wesenheit relations ( Riess et al. 2011 ; Riess et al. 2019 ) . The classic Cepheids discovered by Chen et al. (2019) created the first intuitive 3D map of the Milky Way’s warping precession. RR Lyrae stars are powerful probes of the disk and halo of the Galaxy and are useful to explore the chemical and dynamical evolution of the Galaxy ( Medina et al. 2022 ; Hansen et al. 2011 ; Hansen et al. 2016 ; Li & Binney 2022 ) . By tracing the radial velocity of eclipsing binary stars, physical parameters (such as the radius and mass) of the stars in the system can be derived ( Torres et al. 2010 ) . Pamos Ortega et al. (2023) dated young open clusters (Trumpler 10 and Praesepe) using asteroseismology studies of \delta Scuti stars. Over the past decades, numerous variable star classifiers have been developed to facilitate the rapid identification and classification of variable stars. Kim et al. (2014) selected 28,392 light curves from the EROS-2 LMC database and extracted 22 variability features of light curves to train a random forest classifier. They used EROS B E -band light curves (either raw- or phase-folded) to derive these features, except for J , Q 3-1|B-R and \eta B-R , which needs both B E -and R E -band light curves. Kim & Bailer-Jones (2016) provided a Python package “UPSILoN” containing a pre-trained random forest classifier trained on the light curves of 143,923 periodic variables from the OGLE I-band and EROS-2 B E -band ( Tisserand et al. 2007 ) databases, They choosed 16 features that were relatively survey-independent, i.e. excluded the ones based colors. Xu et al. (2022) selected 241,154 periodic variable stars from ASAS-SN ( Shappee et al. 2014 ; Jayasinghe et al. 2018a ; Jayasinghe et al. 2018b ) V-band and OGLE ( Udalski 2003 ) I-band. They extracted 17 features to train the random forest classifier. Szklenár et al. (2022) created a Multiple-Input Neural Network, consisting of Convolutional and Multi-layer Neural Networks. The network was trained and tested on OGLE-III I-band phase-folded light curves, while taking also into account additional numerical information (e.g. period, reddening-free brightness) to differentiate visually similar light curves. Abdollahi et al. (2023) used the Hierarchical Classification technique and trained the Neural Networks based on the I-band photometry of variables in OGLE-IV database. They folded the light curves by using the specific periods provided by the OGLE catalogue and tried to bin and normalize them. They used 50 values of the binned magnitudes and the period of each light curve as the initial data. All these classifiers are trained by using features extracted from single-band or multi-band light curves, with color-related information sometimes used as additional input. However, building effective variable star classifiers for every band and survey, especially when dealing with large-scale data and complex deep learning models, is a highly resource-intensive process. It involves multiple steps such as visual classification, manual source screening ( Jayasinghe et al. 2018b ) , data cleaning, feature construction, model training, hyperparameter tuning, validation evaluation, and iterative optimization ( Chaini et al. 2024 ; Enis Cody et al. 2024 ) , each requiring substantial computing resources and time. Certain types of variable stars exhibit a degree of consistency in different bands. Cepheid variables show stable Period-Luminosity (PL) relations and Period-Wesenheit (PW) relations in multiple bands. These relations are particularly evident in the optical bands but can also be extended to the near-infrared bands ( Ngeow et al. 2022a ; Ngeow et al. 2022b ; Muñoz et al. 2024 ) . Similarly, RR Lyrae stars demonstrate a consistent relation between their periods and metallicity across different bands ( Bhardwaj 2024 ; Narloch et al. 2024 ) . \delta Scuti stars display similar frequency modes across different bands. Jia et al. (2024) found 2254 multi-mode \delta Scuti stars from the Zwicky Transient Facility Survey. And 1254 objects showed reliable periods in both bands. These findings lead us to consider developing a variable star classifier that is both efficient and resource-saving across multiple bands. To test this idea, we should build a classifier based on photometric data from one band and then compare the classification consistency on photometric data from the second and third bands. However, we do not have suitable data from three bands to perform this work, we instead use photometric data from ASAS-SN V band and ZTF DR2 r band. Since the light curves from the two bands satisfy the requirements for training a robust classifier and can be used to compare the classification consistency, we construct a random forest classifier primarily based on V-band light curves and apply it to classify V-band and r-band light curves. We think the classification differences between V band and r band can already tell us the limitation of the cross-band application of the classifier. We systematically compare the classification results under various preprocessing methods, including the use of magnitude light curves and normalized flux light curves, periods extracted from single-band and multi-band light curves, and the consideration of color features, thereby verifying to some extent whether different bands have varying impacts on variable star classification. We introduce the variable star samples for training the random forest classifier and verifying the multi-band classification differences in Section 2 . We show the classification results of the classifier on variable star samples in different cases in Section 3 . We summarize and compare these results in

Read the complete text