The discrimination between star-forming and AGN galaxies in the absence of H{\alpha} and [NII]: A machine learning approach

TL;DR

This study demonstrates that spectral features like [OIII], Hβ, and the 4000Å-break can effectively distinguish AGN from star-forming galaxies without Hα and [NII], outperforming methods based on stellar mass.

Contribution

It introduces a machine learning approach using spectral features to classify galaxies in the absence of key emission lines, improving accuracy over traditional methods.

Findings

Spectral features can replace missing emission lines for galaxy classification.

The method achieves higher accuracy than stellar mass-based predictions.

Color and physical parameters effectively discriminate galaxy types.

Abstract

In the absence of the two emission lines H$\alpha$ and [NII] (6584\AA) in a BPT diagram, we show that other spectral information is sufficiently informative to distinguish AGN galaxies from star-forming galaxies. We use pattern recognition methods and a sample of galaxy spectra from the Sloan Digital Sky Survey (SDSS) to show that, in this survey, the flux and equivalent width of [OIII] (5007\AA) and H$\beta$, along with the 4000\AA-break, can be used to classify galaxies in a BPT diagram. This method provides a higher accuracy of predictions than those which use stellar mass and [OIII]/H$\beta$. First, we use BPT diagrams and various physical parameters to re-classify the galaxies. Next, using confusion matrices, we determine the `correctly' predicted classes as well as confused cases. In this way, we investigate the effect of each parameter in the confusion matrices and rank the physical parameters used in the discrimination of the different classes. We show that in this survey, for example, $\rm{g - r}$ colour can provide the same accuracy as galaxy stellar mass to predict whether or not a galaxy hosts an AGN. Finally, with the same information, we also rank the parameters involved in the discrimination of Seyfert and LINER galaxies.