Evaluating the feasibility of interpretable machine learning for globular cluster detection

TL;DR

This study assesses machine learning techniques for efficiently detecting extragalactic globular clusters in large photometric datasets, demonstrating high accuracy and transferability across different galaxy environments.

Contribution

It introduces the application of interpretable ML models to globular cluster detection, showing they can match existing catalogues and generalize across datasets.

Findings

ML models recover 90-94% of GCs with 6-8% false positives

High performance (98-99%) in the magnitude range 22-24.5 mag

Models are transferable between different galaxy clusters

Abstract

Extragalactic globular clusters (GCs) are important tracers of galaxy formation and evolution. Obtaining GC catalogues from photometric data involves several steps which will likely become too time-consuming to perform on the large data volumes that are expected from upcoming wide-field imaging projects such as Euclid. In this work, we explore the feasibility of various machine learning (ML) methods to aid the search for GCs. We use archival Hubble Space Telescope data in the F475W and F850LP bands of 141 early-type galaxies in the Fornax and Virgo galaxy clusters. Using existing GC catalogues to label the data, we obtain an extensive data set of 84929 sources containing 18556 GCs and we train several ML methods both on image and tabular data containing physically relevant features extracted from the images. We find that our evaluated ML models are capable of producing catalogues of similar quality as the existing ones. The best performing methods, ensemble-based models like random forests and convolutional neural networks, recover ~ 90-94 % of GCs while producing an acceptable amount of false detections (~ 6-8 %) - with some falsely detected sources being identifiable as GCs that have not been labelled as such in the used catalogues. In the magnitude range 22 < m4_g < 24.5 mag, 98 - 99 % of GCs are recovered. We even find such high performance levels when training on Virgo and evaluating on Fornax data (and vice versa), illustrating that the models are transferable to environments with different conditions such as different distances than in the used training data. Additionally, we demonstrate how interpretable methods can be used to better understand model predictions, recovering that magnitudes, colours, and sizes are important for identifying GCs. These are encouraging results, indicating that similar methods can be applied for creating GC catalogues for a large number of galaxies.