Classification of Fermi-LAT unassociated sources with machine learning in the presence of dataset shifts

TL;DR

This paper applies machine learning to classify Fermi-LAT sources, especially unassociated ones, while addressing dataset shifts that cause differences in source parameter distributions.

Contribution

It introduces a method to handle dataset shifts in classifying Fermi-LAT sources, improving probabilistic classification of unassociated sources.

Findings

Effective classification of unassociated sources achieved

Addressed dataset shift problem in astrophysical data

Enhanced probabilistic source identification

Abstract

About one third of Fermi Large Area Telescope (LAT) sources are unassociated. We perform multi-class classification of Fermi-LAT sources using machine learning with the goal of probabilistic classification of the unassociated sources. A particular attention is paid to the fact that the distributions of associated and unassociated sources are different as functions of source parameters. In this work, we address this problem in the framework of dataset shifts in machine learning.