Machine Learning Techniques for Stellar Light Curve Classification

Abstract

We apply machine learning techniques in an attempt to predict and classify stellar properties from noisy and sparse time-series data. We preprocessed over 94 GB of Kepler light curves from the Mikulski Archive for Space Telescopes (MAST) to classify according to 10 distinct physical properties using both representation learning and feature engineering approaches. Studies using machine learning in the field have been primarily done on simulated data, making our study one of the first to use real light-curve data for machine learning approaches. We tuned our data using previous work with simulated data as a template and achieved mixed results between the two approaches. Representation learning using a long short-term memory recurrent neural network produced no successful predictions, but our work with feature engineering was successful for both classification and regression. In particular, we were able to achieve values for stellar density, stellar radius, and effective temperature with low error (∼2%-4%) and good accuracy (∼75%) for classifying the number of transits for a given star. The results show promise for improvement for both approaches upon using larger data sets with a larger minority class. This work has the potential to provide a foundation for future tools and techniques to aid in the analysis of astrophysical data.

Detecting Exoplanet Transits through Machine-learning Techniques with Convolutional Neural Networks

“…Model 2, called 1D-CNN-folding-0, is a 1D-CNN model that uses whole light curves without folding. Previous work such as Zucker & Giryes (2018) and Hinners et al (2018) employed models like this. The basic structures of our convolutional layers followed Zucker & Giryes (2018), but the input data number is different, and we use three fully connected layers.…”

Classical Machine Learning Techniques in the Search of Extrasolar Planets

“…The recent work of Hinners et al [14] presented different machine learning techniques and models with the objective of classify and predict features over the same data that we use in this paper. Similarly to what we propose, they extracted some statistical features from the light curve, but they were not focused on detecting if the light curve variations were indeed generated by an exoplanet or by other phenomena.…”

“…We used extraction techniques specialized on time series, which in this case corresponds to measurements of light intensity through time, based on Feature Analysis for Time Series (FATS ) library for Python [20]. This library was created with the purpose of extracting characteristics of astronomical time series, originally curves of light, and was previously used in [14] applying FATS over same data we use in this work. Likewise, similar features have been used on other tasks over light curves such as [4,6] did.…”

Variable star classification using multiview metric learning

“…Likewise, standard classification techniques in the astroinformatics-community span a few areas: (i) the classifier is designed such that the user selects features and the classifier is trained on variables with a known type ("expert selected features, for correlation discovery", Debosscher 2009;Sesar et al 2011;Richards et al 2012;Graham et al 2013a;Armstrong et al 2016;Mahabal et al 2017;Hinners et al 2018), (ii) the classifier is designed such that the computer selects the optimal features and the classifier is trained on variables with a known type (McWhirter et al 2017;Naul et al 2018, "computer selected features, for correlation discovery"), (iii) the classifier (clustering algorithm) is designed such that that user selects features and variables with an unknown type are provided ("expert selected features, for class discovery", Valenzuela and Pichara 2018; Modak et al 2018).…”