New results in applying the machine learning to GRB redshift estimation
2017 December 12
Gamma-ray bursts (GRBs) are the most energetic transients in the far Universe. Several thousands of GRBs have been observed so far but we could measure the distance of only a few hundreds. We studied the parameters of GRBs with available spectroscopic redshift in order to be able to estimate the redshift of those GRBs without a measured one. To calculate their distances we applied two machine-learning estimator methods: random forest regressor and XGBoost. For the process we used selected gamma, x-ray and ultraviolet parameters from the Swift GRB catalog, which contains the measured spectroscopic redshift of 328 GRBs. We found a significantly higher correlation between the measured and estimated redshift, we have improved the correlation in multiple steps from 0.57 (published by Ukwatta et al., 2016) to 0.67. It seems that both the random forest and the XGBoost methods give similarly high correlation. For further improvements additional redshift measurements are required.