Preprint of "Machine-Learned Discovery of White Dwarf–Main Sequence Binary Candidates at 100–300 pc with Gaia DR3" (Srinivas 2026). A gradient-boosting classifier trained on the Rebassa-Mansergas et al. (2025, A&A 699, A153) spectroscopic WDMS catalogue identifies 22 UV-confirmed white dwarf–main sequence binary candidates in the solar neighborhood (100–300 pc) using Gaia DR3 BP/RP excess, RUWE, and CMD offset. The classifier recovers 21 of 22 UV-confirmed gold-tier candidates that a recall-matched rectangular cut misses entirely, with a 23-fold improvement in main-sequence contamination rate (0.013% vs 0.29%). Gold-tier selection requires gradient-boosting probability P > 0.8 and Findeisen & Hillenbrand (2010) NUV excess < −1.5 mag, independently confirmed with GALEX photometry. All 22 gold-tier candidates and 368 photometric candidates are available in the accompanying data deposit (Zenodo doi:10.5281/zenodo.20614895).
Akshara Srinivas (Tue,) studied this question.