“APRENDIZADO DE MÁQUINA APLICADO A FOTOMETRIA MULTIBANDA”

Abstract

trophysics, as effective temperature (Teff ), surface gravity (log g), and metallicity (rFe{Hs) are the primary observables that describe stellar structure and evolution. Traditionally, these quantities are derived from spectroscopy, but the growth in the volume of large sur- vey data has motivated the exploration of alternatives that take advantage of their broad coverage. Artificial intelligence (AI) and machine learning (ML) techniques are particu- larly useful in this context, as they can map complex patterns between photometry and physical parameters, enabling precise inferences for large samples. This work develops an ML framework to estimate Teff , log g, and rFe{Hs for stars from the J-PLUS multiband photometric survey, also integrating data from the Gaia and 14/05/2026, 09:36 E-mail de on.br - resumo, abstract, palavras-chave e keywords https://mail.google.com/mail/u/0/?ik=a7e96f56c1&view=pt&search=all&permthid=thread-f:1865100775717255392&simpl=msg-f:18651007757172… 1/2 CatWISE catalogs as well as spectroscopy from the LAMOST survey. The models employ the LightGBM gradient boosting algorithm and are trained with extinction-corrected colors and absolute magnitudes. A systematic feature selection strategy was implemented to identify the most relevant features, totaling 67, 27, and 35 for Teff , log g, and rFe{Hs, respectively. Further reductions are not justified by the resulting information loss, and these were therefore adopted as the final feature sets. The predicted values were compared with those from LAMOST, resulting in competi- tive mean absolute errors of 42 K in Teff , 0.06 dex in log g, and 0.06 dex in rFe{Hs. Applica- tion to a selected sample of 154 members of open clusters and co-moving groups shows that the metallicities of the clusters are generally consistent with those reported in the litera- ture, although the estimates are on average slightly underestimated (´0.12 dex). Beyond the individual analysis of stars, the approach thus provides a uniform and uncertainty- aware AI-based photometric method to estimate cluster metallicities when spectroscopy is not available, enabling guided follow-ups and comparative studies across clusters. Future perspectives include exploring additional combinations of color indices and ex- tending predictions to other relevant parameters, such as rα{Fes, which is fundamental to the study of Galactic chemical evolution. Since the methodology is based on generaliz- able ML principles and on the combination of multiband photometry with astrometry, it can also be applied to similar surveys such as S-PLUS and J-PAS, broadening its future impact and strengthening the role of AI-driven methods in modern stellar astrophysics.