Use of Near-Infrared Colors to Enhance Star-Galaxy Classification in Supervised Learning Models

Abstract

In this paper, we perform star/galaxy separation by using various machine learning models on color information alone. We use a combination of colors obtained from the Hyper Suprime-Cam, the COSMOS Spitzer survey, and the second data release from UltraVISTA. We use supervised learning models to test if using near-infrared colors with HSC colors can perform accurate galaxy/separation—while also minimizing the number of galaxies misclassified as stars. We use an extreme deconvolution model, logistic regression, and support vector machines with linear and radial basis function kernels. Our results suggest that the RBF kernel high-dimensionality is beneficial in performing classification as it outperforms the other models in the distribution of misclassified sources and our metrics. However, the other models have room for improvement as they could be better tuned or include more parameters.