International Conference on Machine Learning for Astrophysics - ML4Astro

QSOs selection in highly unbalanced photometric datasets: The “Michelangelo” reverse-selection method.

Not scheduled

20m

Catania

Oral Presentation Supervised Learning

Speaker

Giorgio Calderone (Istituto Nazionale di Astrofisica (INAF))

Description

I will present a novel selection method aimed at efficiently identifying high-redshift QSOs in highly unbalanced photometric datasets, characterized by a very low number of QSOs with respect to other sources. The method relies on a gradient boosting algorithm, although it may be be used with any other machine learning method providing classification probabilities. I applied the selection method on a sample of photometric data obtained by PanSTARRS1 (DR2), DES (Gold Y3), Gaia (EDR3) and WISE, and I will discuss its performances, as well as a comparison to its basic, direct-selection method counterpart, showing that the former privileges the selection completeness, while the latter privileges the success rate.