Automated classification of eclipsing binary systems in the VVV Survey

TL;DR

This paper develops a supervised machine learning approach to automatically classify eclipsing binary systems in the VVV Survey using light curve data, enhancing large-scale data processing in astronomy.

Contribution

It introduces a novel machine learning model for classifying eclipsing binaries in the VVV Survey, utilizing light curve features and discussing model efficiency and feature importance.

Findings

High classification accuracy achieved for eclipsing binary types

Feature importance analysis identifies key light curve parameters

Model demonstrates potential for large-scale automated classification

Abstract

With the advent of large-scale photometric surveys of the sky, modern science witnesses the dawn of big data astronomy, where automatic handling and discovery are paramount. In this context, classification tasks are among the key capabilities a data reduction pipeline must possess in order to compile reliable datasets, to accomplish data processing with an efficiency level impossible to achieve by means of detailed processing and human intervention. The VISTA Variables of the V\'ia L\'actea Survey, in the southern part of the Galactic disc, comprises multi-epoch photometric data necessary for the potential discovery of variable objects, including eclipsing binary systems (EBs). In this study we use a recently published catalogue of one hundred EBs, classified by fine-tuning theoretical models according to contact, detached or semi-detached classes belonging to the tile d040 of the VVV. We describe the method implemented to obtain a supervised machine learning model, capable of classifying EBs using information extracted from the light curves of variable object candidates in the phase space from tile d078. We also discuss the efficiency of the models, the relative importance of the features and the future prospects to construct an extensive database of EBs in the VVV survey.