Clustering and classification of gamma ray bursts using machine learning techniques

Abstract

Gamma-ray bursts are the most energetic explosions in the universe with a staggering amount of energy release greater than the sun within a fraction of seconds. Although half a decade has passed since its discovery, the physics of GRBs remains enigmatic. As a first step towards unraveling the mysteries of this phenomenon, several classification attempts have been made, but to date none of them have been completely successful. Duration-based classification reveals two distinct classes of GRBs based on their progenators, long and short which are considered to be a byproduct of collapse of massive stars and merger events involving compact objects. But more recently, this classification scheme has been facing a lot of road blocks since the discovery of long GRBs along with kilonovae and short GRBs along with supernovae, which contradicts the traditional classification scheme. The use of state-ofthe- art machine learning techniques are instrumental in addressing this issue. So the idea is to address this anomaly in classification using machine learning techniques such as dimensionality reduction and clustering. Clustering the high-dimensional light curve of the GRBs in a two-dimensional plane and identifying the key astrophysical significance of the grouping which could unravel the mysteries related to the physics of the burst, emission mechanism etc, is the idea behind this work. However unsupervised learning like dimensionality reduction and clustering being sort of a black box, it is never completely understood what parameters in the feature space make them to produce a particular embedding or different clusters. Identifying the key parameters that influence the clustering and answering the most important question on the validity of the clusters we obtained before attributing astrophysical significance is done by considering different methods of analyzing the data and trying to see if the results are consistent enough with the changes introduced and the influence of noise in the analysis is verified by simulating data of different types of noise and carrying out the clustering analysis.