Luminis Stellarum et Machina: Applications of Machine Learning in Light Curve Analysis
Almat Akhmetali, Alisher Zhunuskanov, Aknur Sakan, Marat Zaidyn, Timur Namazbayev, Dana Turlykozhayeva, Nurzhan Ussipov · 2025
Read Luminis Stellarum et Machina: Applications of Machine Learning in Light Curve Analysis by Almat Akhmetali, Alisher Zhunuskanov, Aknur Sakan, Marat Zaidyn, Timur Namazbayev, Dana Turlykozhayeva, Nurzhan Ussipov (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-2504-10038.txt.
Introduction
Almat Akhmetali and co-authors survey, in a 2025 review, how machine learning is used to study light curves, the brightness records of stars and other sky objects over time. Modern telescopes pour out such records faster than classic methods can mine them. Learning algorithms now classify objects, predict behavior, spot patterns, and flag oddballs.
The survey walks through the main methods and their uses, from planet hunts to sorting variable stars and catching supernovae. It also names the field's open wounds: lopsided class counts, noisy or sparse data, hard feature design, and models whose logic is opaque. By mapping what works and what lacks, the review aims to steer the next wave of big-data astronomy.
Catalog record
| CREATOR | Almat Akhmetali, Alisher Zhunuskanov, Aknur Sakan, Marat Zaidyn, Timur Namazbayev, Dana Turlykozhayeva, Nurzhan Ussipov |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 68 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-10038
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Luminis Stellarum et Machina: Applications of Machine Learning in Light Curve Analysis Almat Akhmetali, Alisher Zhunuskanov, Aknur Sakan, Marat Zaidyn, Timur Namazbayev, Dana Turlykozhayeva, Nurzhan Ussipov Abstract The rapid advancement of observational capabilities in astronomy has led to an exponential growth in the volume of light curve (LC) data, presenting both opportunities and challenges for time-domain astronomy. Traditional analytical methods often struggle to fully extract the scientific value of these vast datasets, especially as their complexity increases. Machine learning (ML) algorithms have become indispensable tools for analyzing light curves, offering the ability to classify, predict, discover patterns, and detect anomalies. Despite the growing adoption of ML techniques, challenges remain in LC classification, including class imbalance, noisy data, and interpretability of models. These challenges emphasize the importance of conducting a systematic review of ML algorithms specifically tailored for LC analysis. This survey provides a comprehensive overview of the latest ML techniques, summarizing their principles and applications in key astronomical tasks such as exoplanet detection, variable star classification, and supernova identification. It also discusses strategies to address the existing challenges and advance LC analysis in the near future. As astronomical datasets continue to grow, the integration of ML and deep learning (DL) techniques will be essential for unlocking the full scientific potential of LC data in the era of astronomical big data. I Introduction Time domain astronomy (TDA) is a rapidly evolving field, driven by the continual discovery of new phenomena and the expansion of observational capabilities Ball & Brunner 2010 ; Pesenson et al. 2010 . This field investigates time-varying characteristics of celestial objects using data from diverse messengers, including gravitational waves, neutrinos, and electromagnetic radiation across various photon-energy bands Vaughan 2013 . With the advent of advanced wide-field, multi-epoch sky surveys such as the Sloan Digital Sky Survey (SDSS) York et al. 2000 , Zwicky Transient Facility (ZTF) Graham et al. 2019 , Panoramic Survey Telescope and Rapid Response System (Pan-STARRS) Kaiser et al. 2010 , and the upcoming Vera C. Rubin Observatory (formerly LSST) Ivezić et al. 2019 , the volume of transient discoveries has grown exponentially. For instance, ZTF generates alerts for over 100,000 events each night, while LSST is projected to surpass this by an order of magnitude. These alerts, which indicate significant flux density changes or new spatial positions, represent invaluable data streams for studying transient phenomena such as supernovae, pulsars, and gamma-ray bursts. TDA focuses on analyzing time-series data to construct light curves (LCs), which depict the variation in brightness of celestial objects over time and are fundamental to understanding the physical processes driving these phenomena. However, harnessing these vast data streams requires addressing significant challenges. It is important for astronomers to take into account the possible uncertainties and biases that may arise from observational methods as well as from data analysis procedures. Observational uncertainties stem from limitations in instrumentation and data collection. Analytical uncertainties, on the other hand, emerge from the complexities of modeling physical processes and the numerical methods used to approximate them. Mitigating biases requires developing comprehensive models that incorporate diagnostics from multiple perspectives. This necessitates detailed multi-physics simulations that combine data from various sources and rely on cutting-edge computational capabilities. As a result, TDA not only facilitates the study of astrophysical phenomena but also serves as a bridge to advancing fundamental physics, making it a cornerstone of modern astronomical research. Machine learning (ML) has become an essential tool for automating data classification, archiving, and retrieval Fluke & Jacobs 2020 ; Baron 2019 . Unlike traditional feature extraction methods, which require domain expertise, deep learning (DL) automatically learns data representations. As a subset of ML, DL has gained widespread popularity and has been successfully applied in various domains, including speech recognition, image processing, and natural language processing Nassif et al. 2019 ; Deng & Li 2013 ; Rodellar et al. 2018 ; Mishra & Celebi 2016 ; Ussipov et al. 2024a ; Ussipov et al. 2024b ; Zhao et al. 2019 ; Pathak et al. 2018 ; Ezugwu et al. 2022 ; Ahuja et al. 2020 ; Otter et al. 2020 ; Young et al. 2018 . In astronomy, ML has emerged as a powerful tool for automating the classification and analysis of LCs, enabling researchers to extract insights from vast datasets. Despite the growing adoption of ML techniques, there remains a need for a systematic review of algorithms and models specifically tailored for LC classification, particularly in addressing challenges such as class imbalance, noise, and interpretability. Interpretability is crucial in astronomical applications, as it ensures that ML models not only provide accurate predictions but also offer insights into the underlying physical processes, fostering trust and enabling scientific discovery. This article provides a comprehensive overview of ML algorithms and DL models in LC analysis, covering applications such as exoplanet detection, variable star classification, and supernova identification. By presenting a structured and insightful examination of ML techniques for LC analysis, this review contributes to a deeper understanding of current methodologies and offers guidance for their effective application in future astronomical research. The rest of the paper is organized as follows. Section II provides an overview of major photometric data surveys relevant to LC analysis. Section III introduces fundamental ML concepts and architectures. Section IV explores key applications of ML techniques to LC analysis. Section V outlines current challenges and open research questions in the field. Finally, Section VI concludes with a summary and outlook. II Photometric Survey Datasets Over the past few decades, large-scale photometric surveys have transformed time-domain astronomy, providing vast, multi-wavelength datasets that are readily accessible through online databases. These surveys have been essential in detecting and classifying variable stars, discovering exoplanets, and identifying transient astronomical events. As shown in Figure 1 , the data volume from astronomical surveys has grown exponentially, approximately doubling every 16 months. The combination of different survey datasets enables comprehensive studies of stellar variability across various timescales and wavelengths, driving both traditional astrophysical research and modern ML applications. Figure 1: The total data volume produced by various astronomical surveys throughout their lifetimes. The data reveals that the volume of astronomical survey data doubles approximately every 16 months. The data and figure are sourced from Kremer et al. 2017 ; Smith & Geach 2023 . Photometric surveys can be broadly categorized into ground-based and space-based missions. Ground-based surveys provide wide-field and high-cadence monitoring, often covering extensive regions of the sky over long time periods. In contrast, space-based missions offer high-precision, uninterrupted observations unaffected by atmospheric distortions, enabling the detection of minute variations in stellar brightness. Table 1 provides an overview of major ground-based and space-based photometric surveys, including their active years, observed bands, sky coverage, and references. II.1 Ground-based Observatories Ground-based photometric surveys have significantly expanded understanding of stellar variability, exoplanetary systems, and the dynamic nature of the night sky. Through wide-field imaging and long-term monitoring, these surveys detect brightness variations in stars, uncovering key processes involved in stellar evolution. Covering different sky regions and wavelength bands, they provide extensive datasets that support various astrophysical studies. Below, ground-based observatories are grouped based on their primary target regions. • All-sky : ASAS, ROTSE, WASP, WISE, ASAS-SN • Milky Way (MW) : MACHO, OGLE, UKIDSS • Wide field :