Feature-Based and Deep Learning Methods for Astronomical Light Curve Classification

V. Baymakanov; Kh. Khamitov; A. Raimova; A. Isaeva; R. Taran; M. Zedan; M. Kornilov; M. Demianenko

cs.LGNov 3, 2026

Feature-Based and Deep Learning Methods for Astronomical Light Curve Classification

Authors: V. Baymakanov, Kh. Khamitov, A. Raimova, A. Isaeva, R. Taran, M. Zedan, M. Kornilov, M. Demianenko,

Organizations: HSE University · Southern Federal University · Lomonosov Moscow State University · Max-Planck-Institut für Astronomie ·

Abstract

We compare feature-based machine learning and end-to-end deep learning for variable star classification using the Zwicky Transient Facility (ZTF) StarEmbed dataset, which contains about 40,000 objects across seven imbalanced classes. Classical models trained on physically motivated features consistently outperform deep networks trained directly on raw light curves. We find that limited labeled data contributes to the performance gap, while the multivariate, irregularly sampled, and morphologically diverse nature of light curves poses additional challenges for deep learning models. While deep learning holds promise for learning representations directly from observations, our results indicate that current models fail to generalize with the available training set size. These findings motivate further investigation of the benefits of scaling labeled datasets and self-supervised pretraining on large unlabeled surveys. This work establishes a quantitative baseline for future research and highlights the need for further innovations to unlock the potential of deep learning in astronomical time series analysis.

Figures & tables

Class	EW	RRc	EA	RRab	RS CVn	LPV	RRd	Total
Number of objects	27,949	4,792	4,251	2,068	1,401	457	439	41,357

Table 1: Number of objects per class in the ZTF StarEmbed dataset.

Figure 1: Examples of light curves from the ZTF StarEmbed dataset for the EW, EA, RRc, and LPV classes in the $g$ and $r$ passbands.

Figure 2: Top-9 features with the highest importance scores, as determined by the CatBoost classifier trained on light-curve features.

Figure 3: The best-performing recurrent architecture: input projection, two bidirectional LSTM layers with hidden size 256, masked aggregation, and a classification head.

Figure 4: The deep Transformer architecture: input projection, sinusoidal positional encoding, four pre-norm encoder layers, masked aggregation, and an MLP classification head.

Figure 5: The three augmentations applied to the same light curve of 2,982 observations; the 512 selected observations are shown in blue, the rest in gray.

Architecture	Acc.	Bal. Acc.	Macro F1	ROC-AUC
Extracted Features (Logistic Regression)	$0.787_{\scriptstyle.005}$	$\textbf{0.801}_{\scriptstyle.011}$	$0.661_{\scriptstyle.008}$	$0.955_{\scriptstyle.005}$
Extracted Features (CatBoost)	$\textbf{0.907}_{\scriptstyle.004}$	$0.742_{\scriptstyle.009}$	$0.771_{\scriptstyle.009}$	$\textbf{0.981}_{\scriptstyle.003}$
Extracted Features (MLP)	$0.903_{\scriptstyle.003}$	$0.762_{\scriptstyle.010}$	$\textbf{0.779}_{\scriptstyle.009}$	$\textbf{0.981}_{\scriptstyle.004}$
LSTM (Linear)	$0.609_{\scriptstyle.005}$	$0.653_{\scriptstyle.011}$	$0.508_{\scriptstyle.008}$	$0.914_{\scriptstyle.003}$
LSTM (Linear), Rnd. subs.	$0.437_{\scriptstyle.006}$	$0.505_{\scriptstyle.009}$	$0.336_{\scriptstyle.006}$	$0.833_{\scriptstyle.005}$
LSTM (Linear), Rnd. bins	$0.663_{\scriptstyle.005}$	$0.664_{\scriptstyle.011}$	$0.508_{\scriptstyle.007}$	$0.921_{\scriptstyle.004}$

Table 2: Test-set results of all experiments (standard error as subscript). Best values in bold.

Time Series Classification

Time series classification assigns labels to sequential data, using temporal patterns to support decisions in health, sensing, science, and industry.

See all 185 papers

Domain-Informed Multi-View Self-Distillation for Astronomical Light-Curve Representation Learning with JEPA
Jun 26, 2026Yicheng RuiIrregular Time-Series Modeling Self-Supervised Time Series Representation Learning
Tsung-Dao Lee Institute, Shanghai Jiao Tong University, Shanghai 201210, China
Contrastive Astromer: Two-Stage Contrastive Representation Learning for Astronomical Light Curve Classification
Oct 30, 2026Justin Minggan Wei, Pavlos Protopapas, Guillermo Cabrera-VivesContrastive Learning Self-Supervised Time Series Representation Learning
School of Engineering and Applied Sciences, Harvard University, Cambridge, MA, 02138, USA · Department of Computer Science, Universidad de Concepción, Edmundo Larenas 219, Concepción, Chile
Leveraging Multimodality for Real-Time Classification of Transients and Variables found by the Zwicky Transient Facility
Jun 30, 2026Ved G. Shah, Nabeel Rehemtulla, Adam A. Miller +13Hierarchical Classification Time Series Classification
Department of Physics and Astronomy, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA · Center for Interdisciplinary Exploration and Research in Astrophysics (CIERA), 1800 Sherman Ave, Evanston, IL 60201, USA · NSF-Simons AI Institute for the Sky (SkAI), 172 E. Chestnut St., Chicago, IL 60611, USA +8

Jun 26, 2026J/K move · Enter open · S save

Domain-Informed Multi-View Self-Distillation for Astronomical Light-Curve Representation Learning with JEPA

Yicheng Rui

Tsung-Dao Lee Institute, Shanghai Jiao Tong University, Shanghai 201210, China

Irregular Time-Series Modeling Self-Supervised Time Series Representation Learning Joint-Embedding Predictive Architecture Multi-View Learning Time Series Representation Learning

Open paper AskPDF arXiv

Light curves describe temporal variations in the brightness of celestial objects. Learning robust representations of light curves is essential for large-scale automatic discovery in the dynamic universe, but existing time-series foundation models often struggle with the uneven sampling, complex noise, and wide range of physical timescales that characterize astronomical observations. We propose a domain-informed representation learning framework for irregular astronomical time series with Joint-Embedding predictive architecture (JEPA), combining semantics-preserving views, uncertainty-aware tokenization, and multi-view self-distillation. The encoders are trained with multi-view self-distillation using LeJEPA regularization on the LEAVES dataset and evaluated on the StarEmbed classification benchmark. On StarEmbed, our model outperforms hand-crafted features on 15 of 16 classification metrics. In few-shot linear probing, it achieves macro-F1 scores of 42.56

\pm

7.21 with one sample per class and 63.58

\pm

1.20 with 100 samples per class, consistently improving over hand-crafted features. Beyond variable-star classification, the learned representation supports similarity search, parameter estimation, and photometric zero-point drift detection. We further evaluate cross-domain adaptation on 12 heterogeneous irregular time-series datasets from PYRREGULAR, where the adapted variant matches or exceeds previous state-of-the-art performance on 5 datasets, compared with at most 3 wins by any single prior baseline. These results demonstrate that domain-informed multi-view self-distillation is an effective strategy for learning representations of irregular time series, while also highlighting that successful time-series representation learning requires domain-specific inductive biases rather than a universally optimal architecture.

Abstract

Time Series Classification

Explore similar work

Domain-Informed Multi-View Self-Distillation for Astronomical Light-Curve Representation Learning with JEPA

Contrastive Astromer: Two-Stage Contrastive Representation Learning for Astronomical Light Curve Classification

Leveraging Multimodality for Real-Time Classification of Transients and Variables found by the Zwicky Transient Facility