| Name: Amit Shukla |
| Affiliation: Indian Institute of Technology Indore |
| Conference ID: ASI2026_1297 |
| Title: A Data-Driven Classification of Gamma-Ray Bursts Using Machine Learning and Power Spectral Density |
| Abstract Type: Poster |
| Abstract Category: High Energy Phenomena, Fundamental Physics and Astronomy |
| Author(s) and Co-Author(s) with Affiliation: Harikrishnan R(Indian Institute of Technology Indore, 453552, India), Dr. Amit Shukla(Indian Institute of Technology Indore, 453552, India) |
| Abstract: Gamma-Ray Bursts (GRBs) are intense flashes of gamma radiation arising from catastrophic cosmic events such as stellar collapse and compact object mergers. They are historically divided into short and long classes based on the prompt emission duration (T90), but this binary scheme does not fully capture the diversity revealed by recent observations. In this work, we investigate the structure of GRB prompt emission using a combination of simulations and unsupervised machine learning applied to Fermi–GBM light curves. The analysis is based on Fourier-domain representations of fixed–length segments, which are embedded using UMAP and, where appropriate, clustered with HDBSCAN. Synthetic light curves with controlled power–law power spectral densities (PSDs) show that the method is sensitive to variability properties: single–component signals arrange in a clear sequence from white to very red noise, while mixed–variability signals at fixed duration form largely distinct groups for different combinations of PSD components. When realistic T90 values are taken into account, however, the dominant structure in the embedded space reduces to a short–versus–long separation, with only a smaller population of steep–PSD events remaining distinct. The same behaviour is seen in the Fermi–GBM sample, where duration governs the large–scale geometry of the embedding and a steep–variability subgroup appears as an outlier population. These results provide a simple, simulation–anchored explanation for the persistence of duration–based classification in GRB studies, while also demonstrating that additional variability information is present but largely masked by the effects of differing time scales and signal superposition. They underline both the potential and the limitations of unsupervised methods for GRB taxonomy and point towards future work that combines duration control with variability–based features to obtain a more physically informative picture of the prompt emission. |
