Redshifted broad absorption line quasars found via machine-learned spectral similarity

TL;DR

This paper introduces a machine learning approach using spectral similarity measures to efficiently identify redshifted broad absorption line quasars in SDSS data, discovering 31 new objects and demonstrating the method's effectiveness.

Contribution

It presents a novel spectral similarity-based method for finding rare quasars, outperforming other techniques and enabling discovery with minimal manual inspection.

Findings

31 new RSBAL quasars discovered

Decision tree similarities most effective

Ensemble methods outperform individual measures

Abstract

We report the discovery of 31 new redshifted broad absorption line quasars (RSBALs) from the Sloan Digital Sky Survey (SDSS). The number of previously known such objects is 19. The identification of the new objects was enabled by calculating similarities between quasar spectra in the SDSS. Using these similarities we look for the objects that are similar to the ones in the original sample, visually inspecting only hundreds, out of over 160,000 spectra considered. We compare the performance of several similarity measures, as well as different methods of employing them, in finding the RSBALs. We find that decision tree based similarities recover the most objects, and that an ensemble of methods performs better than any single one. As the similarities are not tailored for the specific problem of finding RSBALs, they could be used for searching for other types of quasars. The similarities and the code for their calculation are available online.