XClass: An Automated Multiwavelength Machine-Learning Pipeline for Classification of Extragalactic X-ray Sources. I. Pipeline Description

Abstract

Most X-ray sources detected in nearby galaxies by the Chandra X-ray Observatory lack astrophysical classifications. We present XClass (X-ray Classifier for extragalactic sources), an end-to-end machine-learning pipeline that classifies extragalactic X-ray point sources into seven classes: AGN, LMXBs, HMXBs, CVs, low-mass and high-mass foreground stars, and supernova remnants. The key challenge is photometric heterogeneity: training sources are predominantly Galactic with photometry from wide-field surveys (PanSTARRS and 2MASS), while extragalactic targets require Hubble Space Telescope (HST) imaging in a disjoint filter system. We address this through a spectral energy distribution (SED) translation that fits class-appropriate spectral models to each training source and convolves the best-fit model through HST filter curves, producing synthetic magnitudes in a common feature space. The classifier uses an asymmetric two-stage Random Forest: Stage 1 separates broad categories (AGN, X-ray binaries, SNRs, stars) and Stage 2 resolves X-ray binaries into LMXBs and HMXBs using an augmented feature vector that includes Stage 1 probabilities. The training set is assembled from ten Galactic catalogs and extragalactic SNR catalogs, cross-matched with the Chandra Source Catalog v2.1. Features include X-ray hardness ratios, SED-translated HST colors, and X-ray-to-optical flux ratios. We restrict the training set to sources with at least one optical magnitude, avoiding imputation artifacts; the pipeline achieves 99.6% accuracy and balanced accuracy of 0.90 on the resulting 11,374-source optical baseline, with excellent calibration (ECE = 0.002). XClass is modular, generalizable to any HST filter configuration, and will be applied to M31 and M33 in a companion paper.