Machine learning sorts nearly a million quasars from DES imaging

Cross-matching DES DR2 photometry with SDSS DR16 spectra produced a training set of 168,738 point-like objects. A K-Nearest Neighbors classifier separated quasars from stars at 0.99 precision with 0.77 recall, and a hybrid decision-tree regressor then assigned photometric redshifts across 0 < z < 3, with a secondary population recovered near z ≈ 4. A stacked outlier classifier was added to suppress catastrophic redshift failures. The final cleaned catalog of 675,683 objects is intended for baryon acoustic oscillation and large-scale structure analyses.