Asteroseismology, by analyzing the natural oscillations of stars, inverts their internal structure and evolutionary state, and is one of the most penetrating research methods in modern stellar physics. Among its many research subjects,Delta Scuti stars (approximately 1.5–2.5 times the mass of the Sun) are known for their rich pulsation patterns and highly dense oscillatory spectra.This has become an important experimental field for asteroseismology. The pulsations of these stars are mainly driven by the opacity (κ) mechanism of the helium ionization region, while their internal convective cores further induce complex processes such as convective overshoot, chemical mixing, and angular momentum redistribution. At the same time, the relatively rapid rotation causes coupling and frequency splitting of oscillation modes, which greatly increases the difficulty of mode recognition and parameter extraction.

In asteroseismic analysis,The frequency corresponding to the highest peak in the power spectrum, the frequency of the maximum oscillation power, and the large frequency interval Δν are particularly important parameters.Δν is extremely sensitive to the average density of a star and is a core indicator for characterizing its overall structure. However, for δ Scuti stars, rapid rotation and multimode aliasing disrupt the originally regular frequency spacing, posing a significant challenge to traditional methods for measuring Δν.

In recent years, the large-scale, high-precision light curve data acquired by the TESS satellite has greatly expanded the research sample of this type of star.However, the data processing is still computationally intensive and relies on experience, and high-precision parameter extraction is still not easy to achieve.Against this backdrop, machine learning offers a new technological path. Compared to traditional methods, ensemble learning can combine the predictions of multiple models, achieving higher accuracy and stability in complex data environments. Methods such as random forests, gradient boosting, and ridge regression have shown great potential in astronomical data analysis in recent years.

Based on this idea, a research team at the University of Warwick in the UK has built a stacked ensemble learning framework.Key asteroseismic parameters of δ Scuti stars can be predicted directly from TESS light curves.The method achieved remarkable results on a sample of 643 stars: the coefficient of determination R² for all target parameters was higher than 0.77, and it showed good generalization ability on 60 stars that were not used in the training. The prediction results were in high agreement with traditional asteroseismic analysis.

The related research findings, titled "Ensemble Machine Learning Approach to Estimate the Asteroseismic Indices for δ Scuti Stars Observed by TESS", have been published in The Astronomical Journal.

Research highlights:

* A machine learning framework is proposed to directly estimate key asteroseismic parameters from light curves, which breaks through the limitations of traditional methods and significantly improves the efficiency of parameter extraction.

* High-precision predictions were achieved by optimizing feature selection and model architecture, and their reliability was verified on independent samples.

* The asteroseismic index determination of 251 Delta Scuti stars was completed, a new star catalog was constructed, and the parameter database of related stars was enriched, providing important data support for future large-sample statistical analysis and stellar evolution research.

Paper address:
https://beta.iopscience.iop.org/article/10.3847/1538-3881/ae4bd8

More cutting-edge AI papers:

https://hyper.ai/cn/papers

Dataset: TESS light curve screening and asteroseismic sample construction

The core dataset used in this study contains TESS light curves of 643 Delta Scuti stars.In addition, three key asteroseismic indices were included: ν(Aₘₐₓ), νₘₐₓ, and Δν. The initial sample contained 677 δ Scuti stars, of which 643 were retained as the core dataset after multiple rounds of selection. The selection criteria included: possession of TESS 2-minute short-exposure light curves (from the MAST archive); no fewer than 7,000 data points per observation area; light curves corrected by PDC-SAP; and complete and usable asteroseismic parameters.

Building upon this, the researchers selected an additional 251 Delta Scuti stars as supplementary samples. These stars also possess high-quality light curves, but their corresponding asteroseismic parameters have not yet been published. The selection criteria were: coverage of at least three observational regions, with each region containing no fewer than 7,000 data points. This sample was primarily used for practical predictions and validation of the model.

Model: An ensemble regression framework based on stacked multi-base models

The model in this study aims to estimate the asteroseismic parameters of stars based on the characteristics of their light curves.The overall process includes feature extraction, data preprocessing, ensemble modeling, and hyperparameter optimization.

In terms of feature construction,The study employed two types of features: one type is statistical features (such as mean, standard deviation, median, etc.) used to describe the basic properties of luminosity distribution; the other type is frequency domain features, including principal component analysis (PCA), autocorrelation function (ACF), fast Fourier transform (FFT) and discrete wavelet transform (DWT), used to extract periodic and multi-scale structural information in oscillating signals.

In the data preprocessing stage,First, samples with missing values are removed, and the features are normalized. Furthermore, to address the issue of imbalanced feature distribution, a statistical distribution-based resampling method is introduced to generate synthetic data and mitigate bias, thereby improving the stability of model training.

In terms of framework, the model adopts a stacked ensemble regression framework, with random forest, gradient boosting regression, and ridge regression as base models: the first two improve prediction performance by reducing variance and bias, respectively, while ridge regression addresses the collinearity problem among features through regularization. The outputs of the base models are further used as inputs to train a meta-regressor for fusion, thereby improving the overall generalization ability and reducing prediction error.

During model training, researchers also used random search combined with cross-validation to optimize key hyperparameters (such as the number of trees, maximum depth, and learning rate) to obtain a stable and high-performance model configuration.

Generalization was tested using 60 individual stars, and all asteroseismic indices R² > 0.77.

Experimental validation includes three parts: model training, generalization ability evaluation, and prediction of new samples.

During the training phase, researchers randomly selected 583 stars from 643 stars for model building, dividing them into training and test sets at an 8:2 ratio, repeating this process 100 times to reduce the impact of randomness. The remaining 60 stars served as an independent test set to evaluate the model's generalization ability. In addition, 251 unlabeled samples were used for the final prediction.

On training and test samples,The model's prediction R² for ν(Aₘₐₓ), νₘₐₓ, and Δν are 0.95, 0.93, and 0.87, respectively, with the relative error for most samples being less than 0.2.Feature importance analysis shows that the autocorrelation function (ACF) contributes the most, followed by the FFT and DWT. Some statistical features (such as skewness and kurtosis) also play a role. The learning curves show that the model converges and stabilizes, and hyperparameter optimization is effective.

On independent test sets, the model maintained good performance, with R² values of 0.91, 0.87, and 0.77 for the three parameters, respectively, showing high consistency between the predicted results and observed values. The results of repeated experiments showed minimal fluctuations, indicating good stability and robustness of the model. Finally, the researchers applied the model to 251 unlabeled stars, obtaining predicted asteroseismic parameters. The results generally fell within the reasonable range for Delta Scuti stars.

Conclusion

Overall, this work is not a replacement for traditional asteroseismological methods, but rather a targeted supplement: in the context of the rapid accumulation of large-scale observational data, it achieves efficient parameter prediction through data-driven methods, and then combines this with detailed physical modeling for in-depth analysis. This approach is particularly relevant for targets with complex oscillation modes, such as Delta Scuti, which are difficult to standardize.