An Enhanced Catalog of Gaia DR3 Galaxy Candidates with Spectroscopic and Machine-Learning Photometric Redshifts

We present an updated galaxy catalog based on the Gaia Data Release 3 (DR3) Galaxy Candidates table, augmented with both spectroscopic and machine-learning based photometric redshift estimates. While the original catalog of Gaia DR3 galaxy candidates provides an unprecedented all-sky sample of extragalactic sources, its scientific utility for extragalactic and cosmological studies has been limited by the absence of redshift information and by its relatively low source purity. To address these limitations, we cross-match the catalog of Gaia DR3 galaxy candidates with existing spectroscopic redshift catalogs including DESI, SDSS, 2MRS, and NED. For sources without spectroscopic measurements, we compute photometric redshifts using a Multiple-Bin Regression Neural Network (MBRNN) machine-learning model. The input features for this model are drawn from a newly constructed multiwavelength catalog that combines Gaia DR3 measurements with 2MASS and unWISE photometry, as well as shape parameters, QSO classification flags, and extinction values. The model is trained and validated using the spectroscopic subsample. As a result, spectroscopic redshifts are assigned to 24.5\% of the catalog sources, while photometric redshifts are provided for the remaining objects. The MBRNN model achieves improved redshift accuracy and precision compared to previously published photometric redshift estimates based on Gaia data alone. In addition to redshift values, we provide a score quantifying the likelihood that a given source represents a statistical outlier. The resulting catalog significantly enhances the scientific value of the Gaia DR3 galaxy candidates sample and provides a valuable resource for large-scale extragalactic and cosmological studies.