AI and the World's Hard Problems · Astronomy & space · arXiv · 2026-10-20
When X-ray Features Fail to Identify Intrinsic Emitters: Label Noise and Luminosity Overlap in Machine Learning Classification of AGN and Star-forming Galaxies
Adding an X-ray flux feature to a Random Forest classifier of AGN and star-forming galaxies lowered accuracy from 97.51% to 89.26%. The authors trace this to substantial overlap in X-ray luminosity across the 10^42 erg/s threshold and to instance-dependent label noise near the BPT demarcation line. On a fixed sample, adding the X-ray feature left class-resolved metrics unchanged, and its permutation importance was consistent with zero, so the classifier makes negligible use of it.