Target Selection for the Redshift-Limited WAVES-Wide with Machine Learning

TL;DR

This paper presents a machine learning approach using XGBoost to predict galaxy redshift eligibility for the WAVES-Wide survey, aiming to meet its 95% completeness goal without prior redshift knowledge.

Contribution

The study introduces a supervised machine learning method to select target galaxies based on predicted redshift probability, improving survey efficiency.

Findings

Achieved ~95% completeness in target selection.

Demonstrated effectiveness of XGBoost classifier with multi-band photometry.

Validated approach using extensive spectroscopic calibration data.

Abstract

The forthcoming Wide Area Vista Extragalactic Survey (WAVES) on the 4-metre Multi-Object Spectroscopic Telescope (4MOST) has a key science goal of probing the halo mass function to lower limits than possible with previous surveys. For that purpose, in its Wide component, galaxies targetted by WAVES will be flux-limited to $Z<21.1$ mag and will cover the redshift range of $z<0.2$, at a spectroscopic success rate of $\sim95\%$. Meeting this completeness requirement, when the redshift is unknown a priori, is a challenge. We solve this problem with supervised machine learning to predict the probability of a galaxy falling within the WAVES-Wide redshift limit, rather than estimate each object's redshift. This is done by training an XGBoost tree-based classifier to decide if a galaxy should be a target or not. Our photometric data come from 9-band VST+VISTA observations, including KiDS+VIKING surveys. The redshift labels for calibration are derived from an extensive spectroscopic sample overlapping with KiDS and ancillary fields. Our current results indicate that with our approach, we should be able to achieve the completeness of $\sim95\%$, which is the WAVES success criterion.