Multi-wavelength Classification of Active and Star-forming Galaxies on the BPT Diagram with Supervised Machine Learning Models

Distinguishing between active galactic nuclei (AGN) and star-forming galaxies (SFGs) in large astronomical surveys is an important challenge with applications to various subfields of astrophysics and cosmology. In particular, this distinction is especially crucial for informing theories of supermassive black hole formation, which are tied to our understanding of the Big Bang and the early Universe. The Baldwin-Phillips-Terlevich (BPT) diagram, which uses flux ratios of optical emission lines, has been the traditional gold standard for differentiating between these galaxies for nearly four decades. Our aim is to investigate the effect that each spectral line and photometry band has on the accuracy of supervised machine learning models in distinguishing between AGN and SFGs. The dataset we employ consists of nearly thirty thousand galaxies in a crossmatch of data from the Sloan Digital Sky Survey Data Release 16 (SDSS DR16) with ultraviolet, X-ray, and infrared surveys. We find that the addition of multi-wavelength data improves upon both the accuracy and precision of the original BPT diagram, increasing the accuracy to approximately 98%; that machine learning models are able to achieve above 90% accuracy using only photometry, which is much easier and more efficient to collect than spectra. We also find that adding in X-ray flux data does not improve the models, which could be an indication that the X-ray properties of AGN and SFGs are more similar than what was previously theorized.