Photometric selection and redshifts for quasars in the Kilo-Degree Survey Data Release 4

We present a catalog of quasars, their photometric redshifts, and redshift uncertainties derived from the Kilo-Degree Survey (KiDS) Data Release 4. We train machine learning (ML) models using optical ugri and near-infrared ZYJHKs bands, on objects known from SDSS spectroscopy. In 45 million objects of the KiDS data limited to 9-band detections, we define inference subsets based on a feature space built from magnitudes, their combinations, and shape classifiers. We show that projections of a feature space on two dimensions can be successfully used instead of the standard color-color plots, to support the process of building a catalog. The model testing employs two subsets of objects: randomly selected and the faintest ones, which allows us to properly fit the bias vs. variance trade-off. We test three ML models and find that XGBoost is the most robust for classification, while Artificial Neural Networks (ANN) are the best for combined classification and redshift. The catalog is tested using number counts and Gaia parallaxes. Based on these tests, we calibrate the purity vs. completeness trade-off with minimum classification probability for quasar candidates: p(QSO_cand)>0.9 for the safe inference subset at r0.98 for the reliable extrapolation at 22