Robust Assessment of Clustering Methods for Fast Radio Transient Candidates

Fast radio transient search algorithms identify signals of interest by iterating and applying a threshold on set matched filters. These filters are defined properties the such as time dispersion. A real can trigger hundreds trials, each which has to be post-processed for visualization classification tasks. In this paper, we have explored range unsupervised clustering cluster these redundant candidate detections. We demonstrate Realfast, commensal fast-transient system at Karl G. Jansky Very Large Array. use four features clustering: sky position (l, m), time, dispersion measure (DM). develop custom performance metric that makes sure candidates clustered into small number pure clusters, i.e., clusters with either astrophysical or noise candidates. then compare eight different algorithms. show using location along DM/time improves ∼10% compared traditional DM/time-based clustering. Therefore, positional information should used during if it made available. conduct several tests generalizability other data sets propose strategy choose an algorithm. Our easily extended single-pulse pipelines astronomy non-astronomy-based applications.