2019ApJS..242...10S

We used data from the QUEST-La Silla Active Galactic Nucleus (AGN) variability survey to construct light curves for 208,583 sources over ∼70 deg², with a limiting magnitude r ∼ 21. Each light curve has at least 40 epochs and a length of >=200 days. We implemented a random forest algorithm to classify our objects as either AGN or non-AGN according to their variability features and optical colors, excluding morphology cuts. We tested three classifiers, one that only includes variability features (RF1), one that includes variability features and also r - i and i - z colors (RF2), and one that includes variability features and also g - r, r - i, and i - z colors (RF3). We obtained a sample of high-probability candidates (hp-AGN) for each classifier, with 5941 candidates for RF1, 5252 candidates for RF2, and 4482 candidates for RF3. We divided each sample according to their g - r colors, defining blue (g - r <= 0.6) and red subsamples (g - r > 0.6). We find that most of the candidates known from the literature belong to the blue subsample, which is not necessarily surprising given that, unlike many literature studies, we do not cut our sample to point-like objects. This means that we can select AGNs that have a significant contribution from redshifted starlight in their host galaxies. In order to test the efficiency of our technique, we performed spectroscopic follow-up, confirming the AGN nature of 44 among 54 observed sources (81.5% efficiency). From the campaign, we concluded that RF2 provides the purest sample of AGN candidates.