Identification of Extended Emission Gamma-Ray Burst Candidates Using Machine Learning

Abstract

Gamma-ray bursts (GRBs) have been classified traditionally based on their duration. The increasing number of extended emission (EE) GRBs, lasting typically more than 2s but with properties similar to those of short GRBs, challenges the traditional classification criteria. In this work, we use the t-distributed stochastic neighbor embedding (t-SNE), a machine-learning technique, to classify GRBs. We present the results for GRBs observed until 2022 July by the Swift/BAT (Burst Alert Telescope) instrument in all its energy bands. We show the effects of varying the learning rate and perplexity parameters as well as the benefit of preprocessing the data by a nonparametric noise-reduction technique. Consistently with previous works, we show that the t-SNE method separates GRBs into two subgroups. We also show that EE GRBs reported by various authors under different criteria tend to cluster in a few regions of our t-SNE maps and identify seven new EE GRB candidates by using the gamma-ray data provided by the automatic pipeline of Swift/BAT and the proximity with previously identified EE GRBs.

New detailed characterization of the residual luminescence emitted by the GAGG:Ce scintillator crystals for the HERMES Pathfinder mission

“…Even in [107] a t-SNE is employed showing how its hyperparameters learning rate and perplexity influence the results. In the end the method show a good separation in two clusters.…”

Exploring Gamma-Ray Burst Diversity: Clustering Analysis of the Emission Characteristics of Fermi- and BATSE-detected Gamma-Ray Bursts

“…These attempts were made to identify classes in GRBs based on the prompt emission parameters such as duration, hardness ratio, fluences, etc. (Rajaniemi & Mahonen 2002;Hakkila et al 2003) as well as using the morphology of the GRB prompt emission light curves (Jespersen et al 2020;Dimple et al 2023;Garcia-Cifuentes et al 2023;Steinhardt et al 2023), using various machine-learning techniques of unsupervised clustering. Furthermore, we note that parameters such as redshift, kilonova detection, supernova detection, host galaxy information, burst energetics, afterglow properties, and gravitational wave detections, etc.…”