Photometric classification of quasars from ALHAMBRA survey using random forest

Abstract

Context. Given the current era of big data in astronomy, machine-learning-based methods have begun to be applied over recent years to identify or classify objects, such as quasars, galaxies, and stars, from full-sky photometric surveys. Aims. Here we systematically evaluate the performance of random forests (RFs) in classifying quasars using either magnitudes or colours – both from broad- and narrow-band filters – as features. Methods. The working data consist of photometry from the ALHAMBRA Gold Catalogue, which we cross-matched with the Sloan Digital Sky Survey (SDSS) and the Million Quasars Catalogue (Milliquas) for objects labelled as quasars, galaxies, or stars. An RF classifier is trained and tested to evaluate the effects of varying the free parameters and using narrow or broad-band magnitudes or colours on final accuracy and precision. Results. Best performances of the classifier yielded global accuracy and quasar precision of around 0.9. Varying free model parameters (within reasonable ranges of values) has no significant effects on the final classification. Using colours instead of magnitudes as features results in better performances of the classifier, especially when using colours from the ALHAMBRA survey. Colours that contribute the most to the classification are those containing the near-infrared JHK bands.