Investigating machine learning approaches for differentiating MGFs and SGRBs

Abstract

Magnetar giant flares (MGFs) and short gamma-ray bursts (SGRBs) are short gamma-ray transients (SGRTs) with overlapping temporal and spectral characteristics, making them challenging to distinguish, especially when their redshift is unknown. In this study, we apply supervised machine learning using a Support Vector Machine to classify MGFs and SGRBs. Temporal parameters (including pulse rise times from Norris function fits) and spectral features (derived from Comptonized model fits over the 10 keV-40 MeV range) are extracted as input features for classification. We analyse 15 MGF and 101 SGRB samples from 10 Fermi-GBM sources. Classifier performance is assessed using leave-one-source-out cross-validation. The model successfully classifies most SGRBs, but struggles with MGFs due to limited training data and significant feature overlap. This work highlights the potential and challenges of incorporating machine learning into the automated classification of these SGRTs.