The LSST AGN Data Challenge: Selection methods

TL;DR

This paper evaluates machine learning methods for selecting active galactic nuclei (AGN) in large surveys like LSST, demonstrating high accuracy and the importance of variability features for improved classification.

Contribution

It presents a comprehensive comparison of classical and machine learning techniques for AGN classification using simulated LSST data, highlighting the effectiveness of variability features.

Findings

Supervised models achieved 97.5% accuracy.

Unsupervised clustering reached 96.0% accuracy.

Variability features significantly enhance classification performance.

Abstract

Development of the Rubin Observatory Legacy Survey of Space and Time (LSST) includes a series of Data Challenges (DC) arranged by various LSST Scientific Collaborations (SC) that are taking place during the projects preoperational phase. The AGN Science Collaboration Data Challenge (AGNSCDC) is a partial prototype of the expected LSST AGN data, aimed at validating machine learning approaches for AGN selection and characterization in large surveys like LSST. The AGNSC-DC took part in 2021 focusing on accuracy, robustness, and scalability. The training and the blinded datasets were constructed to mimic the future LSST release catalogs using the data from the Sloan Digital Sky Survey Stripe 82 region and the XMM-Newton Large Scale Structure Survey region. Data features were divided into astrometry, photometry, color, morphology, redshift and class label with the addition of variability features and images. We present the results of four DC submitted solutions using both classical and machine learning methods. We systematically test the performance of supervised (support vector machine, random forest, extreme gradient boosting, artificial neural network, convolutional neural network) and unsupervised (deep embedding clustering) models when applied to the problem of classifying/clustering sources as stars, galaxies or AGNs. We obtained classification accuracy 97.5% for supervised and clustering accuracy 96.0% for unsupervised models and 95.0% with a classic approach for a blinded dataset. We find that variability features significantly improve the accuracy of the trained models and correlation analysis among different bands enables a fast and inexpensive first order selection of quasar candidates