ClassiPyGRB: Machine Learning-Based Classification and Visualization of Gamma Ray Bursts using t-SNE

Abstract

Gamma-ray burst (GRBs) are the brightest events in the universe. For decades, astrophysicists have known about their cosmological nature. Every year, space missions such as Fermi and SWIFT detect hundreds of them. In spite of this large sample, GRBs show a complex taxonomy in the first seconds after their appearance, which makes it very difficult to find similarities between them using conventional techniques. It is known that GRBs originate from the death of a massive star or from the merger of two compact objects. GRB classification is typically based on the duration of the burst (Kouveliotou et al., 1993). Nevertheless, events such as GRB 211211A (Yang et al., 2022), whose duration of about 50 seconds lies in the group of long GRBs, has challenged this categorization by the evidence of features related with the short GRB population (the kilonova emission and the properties of its host galaxy). Therefore, a classification based only on their gamma-ray duration does not provide a completely reliable determination of the progenitor. Motivated by this problem, Jespersen et al. (2020) and Steinhardt et al. (2023) carried out analysis of GRB light curves by using the t-SNE algorithm, showing that Swift/BAT GRBs database, consisting of light curves in four energy bands (15-25 keV, 25-50 keV, 50-100 keV, 100-350 keV), clusters into two groups corresponding with the typical long/short classification. However, in this case, this classification is based on the information provided by their gamma-ray emission light curves. ClassiPyGRB is a Python 3 package to download, process, visualize and classify GRBs database from the Swift/BAT Instrument (up to July 2022). It is distributed over the GNU General Public License Version 2 (1991). We also included a noise-reduction and an interpolation tools for achieving a deeper analysis of the data.