Classification Problem and Parameter Estimating of Gamma-Ray Bursts.

There are at least two distinct classes of Gamma-Ray Bursts (GRB) according to their progenitors: short duration and long duration bursts. It was shown that short bursts result from compact binary merging, while long bursts are associated with core collapse supernova. However, one could suspect the existence of more classes and subclasses. For example, compact binary can be double neutron stars, or neutron star and black hole, which might generate gamma-ray bursts with different properties. From another hand, gamma-ray transients are known to be produced by magnetars in Galaxy, named Soft Gamma-Repeaters (SGR). A Giant Flare from SGR can be detected from a nearby galaxy, and it can mimic for a short GRB. So the classification problem is very important for correct investigation of different transient progenitors. Gamma-ray transients are characterized by a number of parameters and known phenomenology correlations between them, obtained for well classified ones. Using these correlations, which could be unique for different classes of gamma-ray transients, we can classify an event and determine the type of its progenitor, using only temporal and spectral characteristics. We suggest the statistical classification method, based on the cluster analysis of the trained dataset of 323 events. Using the known dependencies, one can not only classify the types of gamma-ray bursts, but also discriminate events that are not associated with gamma-ray bursts, but have a different physical nature. We show that GRB 200415A, originally classified as a short GRB, probably does not belong to the class of short GRBs, but it is most likely associated with the giant flare of SGR. On the other hand, we can estimate one of the unknown parameters if we assume the certain classification of the event. As an example, an estimate the redshift of the GRB 200422A source is given. We also discuss that in some cases it is possible to give a probabilistic estimate of the unknown parameters of the source. The method could be applied to any other analogous classification problems.