From seagull to hummingbird: New diagnostic methods for resolving galaxy activity

TL;DR

This paper introduces a highly accurate machine learning diagnostic tool that classifies galaxy activity types and resolves mixed-activity galaxies into their primary excitation mechanisms using spectral features.

Contribution

The study develops a novel random forest-based diagnostic method that surpasses current techniques in classifying galaxy activity and deconstructing mixed-activity galaxies.

Findings

Achieves ~99% overall accuracy in galaxy classification.

Effectively decomposes mixed-activity galaxies into primary excitation sources.

Simplified diagnostic using D4000 index and EW([O iii]) with minimal power loss.

Abstract

Context. A major challenge in astrophysics is classifying galaxies by their activity. Current methods often require multiple diagnostics to capture the full range of galactic activity. Furthermore, overlapping excitation sources with similar observational signatures complicate the analysis of a galaxy's activity. Aims. This study aims to create an activity diagnostic tool that overcomes the limitations of current emission line diagnostics by identifying the underlying excitation mechanisms in mixed-activity galaxies (e.g., star formation, active nucleus, or old stellar populations) and determining the dominant ones. Methods. We use the random forest machine-learning algorithm, trained on three main activity classes -- star-forming, AGN, and passive -- that represent key gas excitation mechanisms. This diagnostic employs four distinguishing features: the equivalent widths of [O iii] ${\lambda}$5007, [N ii] ${\lambda}$6584, H${\alpha}$, and the D4000 continuum break index. Results. The classifier achieves near-perfect performance, with an overall accuracy of ~ 99% and recall scores of ~ 100% for star-forming, ~ 98% for AGN, and ~ 99% for passive galaxies. These exceptional scores allow for confident decomposition of mixed activity classes into the primary gas excitation mechanisms, overcoming the limitations of current classification methods. Additionally, the classifier can be simplified to a two-dimensional diagnostic using the D4000 index and log$_{10}$(EW([O iii])$^{2}$) without significant loss of diagnostic power. Conclusions. We present a diagnostic for classifying galaxies by their primary gas excitation mechanisms and deconstructing the activity of mixed-activity galaxies into these components. This method covers the full range of galaxy activity. Aditionally, D4000 index serves as an indicator for resolving the degeneracy among various activity components.