# Identifying the physical origin of gamma-ray bursts with supervised machine learning

**Authors:** Jia-Wei Luo, Fei-Fei Wang, Jia-Ming Zhu-Ge, Ye Li, Yuan-Chuan Zou, Bing Zhang  
**arXiv:** [2211.16451](https://arxiv.org/abs/2211.16451)  
**Date:** 2022-11-29  
**Summary page:** https://www.summarizepaper.com/en/paper/2211.16451v1/

## Key points

- Gamma-ray bursts (GRBs) are classified into two types: long and short, based on their durations.
- Type I GRBs originate from compact binary mergers, while Type II GRBs originate from massive star collapses.
- There is a significant overlap in the duration distributions of long and short GRBs, and some intermingled GRBs have been reported.
- A multi-wavelength, multi-parameter classification scheme of GRBs is needed to address this issue.
- Jia-Wei Luo et al. build such a classification scheme using supervised machine learning methods, chiefly XGBoost.
- The authors utilize the GRB Big Table and Greiner's GRB catalog and divide the input features into three subgroups: prompt emission, afterglow, and host galaxy.
- The prompt emission subgroup performs the best in distinguishing between Type I and II GRBs.
- The most important distinguishing features in prompt emission are $T_{90}$ (the time interval during which 90% of the burst's fluence is emitted), hardness ratio (the ratio of counts in high-energy to low-energy channels), and fluence (the total energy emitted).
- After building the machine learning model, the authors apply it to currently unclassified GRBs to predict their probabilities of being either Type I or Type II.
- They assign each GRB its most probable physical class based on these probabilities.
- This work provides an effective method for identifying the physical origin of previously unclassified GRBs.
- It highlights the importance of considering multiple parameters when classifying these events.
- The refined classification scheme developed in this study could aid future research on understanding the astrophysical processes behind these powerful explosions by providing an efficient way to classify them accurately according to their physical origins.

## Summary

Gamma-ray bursts (GRBs) are classified into two types, long and short, based on their durations. This classification is linked to the physical origin of GRBs, with Type I GRBs originating from compact binary mergers and Type II GRBs originating from massive star collapses. However, there is a significant overlap in the duration distributions of long and short GRBs, and some intermingled GRBs have been reported. To address this issue, a multi-wavelength, multi-parameter classification scheme of GRBs is needed. In this paper by Jia-Wei Luo et al., the authors build such a classification scheme using supervised machine learning methods, chiefly XGBoost. The authors utilize the GRB Big Table and Greiner's GRB catalog and divide the input features into three subgroups: prompt emission, afterglow, and host galaxy. They find that the prompt emission subgroup performs the best in distinguishing between Type I and II GRBs. The most important distinguishing features in prompt emission are $T_{90}$ (the time interval during which 90% of the burst's fluence is emitted), hardness ratio (the ratio of counts in high-energy to low-energy channels), and fluence (the total energy emitted). After building the machine learning model, the authors apply it to currently unclassified GRBs to predict their probabilities of being either Type I or Type II. They assign each GRB its most probable physical class based on these probabilities. The authors' work provides an effective method for identifying the physical origin of previously unclassified GRBs. It also highlights the importance of considering multiple parameters when classifying these events. The refined classification scheme developed in this study could aid future research on understanding the astrophysical processes behind these powerful explosions by providing an efficient way to classify them accurately according to their physical origins.