QUAS(AI)R: A Novel Machine Learning Algorithm to Predict X-ray Brightness in Active Galactic Nuclei

Abstract

Active Galactic Nuclei (AGNs) are a compact region at the center of galaxies that emit more energy than the rest of the galaxy itself. They emit light across the electromagnetic spectrum, from radio waves to optical light to high-energy X-rays. AGNs indicate the existence of highly energetic phenomena in the nucleus of the galaxy. Although AGNs were identified 60 years ago, our knowledge about their physical properties is limited. Quasars, a subtype of AGNs, provide some of the most intense forms of X-rays, which are among the most energetic light known. Furthering our understanding of X-rays in the dynamic environments of quasars will add to our understanding of how to benefit from their use on Earth. In my project, I study the X-ray brightness in quasars to develop six types of regression-based machine learning models for the X-ray brightness predictions. These six models were Stochastic Gradient Descent (SGD), Random Forest, Ridge, Lasso, Bayesian and the baseline linear regression model, built on the scikit-learn Python package. The training/testing split on the MILLIQUAS dataset was 80/20 percent, and each model was tuned on model-specific hyperparameters. Benchmarked with the normalized mean absolute error (NMAE), the top three performing models were the Bayesian (0.022%), Ridge (0.180%), and Lasso (0.183%), with the baseline NMAE at 0.284%. With this, we can learn more about the evolution of galaxies in the early Universe and understand how these dynamic environments came to be.