# Probing the unidentified Fermi blazar-like population using optical polarization and machine learning

## TL;DR

This study combines optical polarization data and machine learning to identify blazar-like sources among Fermi gamma-ray unidentified objects, achieving over 95% accuracy and successfully confirming candidate sources.

## Contribution

It introduces a novel approach integrating optical polarization and machine learning to classify gamma-ray sources, improving identification accuracy of blazar counterparts.

## Key findings

- Achieved >95% classification accuracy in identifying blazar candidates.
- Optical polarization significantly enhances the machine learning identification process.
- Successfully confirmed a candidate blazar source in an unidentified Fermi field.

## Abstract

The Fermi gamma-ray space telescope has revolutionized our view of the gamma-ray sky and the high energy processes in the Universe. While the number of known gamma-ray emitters has increased by orders of magnitude since the launch of Fermi, there is an ever increasing number of, now more than a thousand, detected point sources whose low-energy counterpart is to this day unknown. To address this problem, we combined optical polarization measurements from the RoboPol survey as well as other discriminants of blazars from publicly available all-sky surveys in machine learning (random forest and logistic regression) frameworks that could be used to identify blazars in the Fermi unidentified fields with an accuracy of >95%. Out of the potential observational biases considered, blazar variability seems to have the most significant effect reducing the predictive power of the frameworks to ~80-85%. We apply our machine learning framework to six unidentified Fermi fields observed using the RoboPol polarimeter. We identified the same candidate source proposed by Mandarakas et al. for 3FGL J0221.2+2518.