Machine learning classification of repeating FRBs from FRB 121102

Abstract

ABSTRACT Fast radio bursts (FRBs) are mysterious bursts in the millisecond time-scale at radio wavelengths. Currently, there is little understanding about the classification of repeating FRBs, based on difference in physics, which is of great importance in understanding their origin. Recent works from the literature focus on using specific parameters to classify FRBs to draw inferences on the possible physical mechanisms or properties of these FRB subtypes. In this study, we use publicly available 1652 repeating FRBs from FRB 121102 detected with the Five-hundred-metre Aperture Spherical Telescope (FAST), and studied them with an unsupervised machine learning model. By fine-tuning the hyperparameters of the model, we found that there is an indication for four clusters from the bursts of FRB 121102 instead of the two clusters (‘Classical’ and ‘Atypical’) suggested in the literature. Wherein, the ‘Atypical’ cluster can be further classified into three sub-clusters with distinct characteristics. Our findings show that the clustering result we obtained is more comprehensive not only because our study produced results which are consistent with those in the literature but also because our work uses more physical parameters to create these clusters. Overall, our methods and analyses produced a more holistic approach in clustering the repeating FRBs of FRB 121102.