Machine learning-based prediction of nitrogen-fixing efficiency in Cowpea rhizobia from the Brazilian semiarid.

This study explores the potential of machine learning to predict nitrogen fixation efficiency in rhizobia strains associated with cowpea (Vigna unguiculata), aiming to optimize bioinoculant selection for sustainable agriculture. Eight native strains were isolated from soils in the Brejo Paraibano region (Brazil), characterized morphologically on Yeast Mannitol Agar YMA medium, and evaluated in greenhouse bioassays for nitrogen accumulation and Relative Index of Nitrogen Fixation Efficiency (IRF%). A Ridge Regression model was then developed using phenotypic colony traits as predictors to estimate Total Nitrogen and IRF%. The results demonstrated strong correlations between predicted and actual values (r = 0.95-0.96), suggesting that visible colony characteristics can serve as reliable proxies for strain efficiency. This approach has the potential to offer a cost-effective alternative to traditional greenhouse trials, with indications of reduced time and resource demands. However, these results are theoretical and require validation through larger datasets and field conditions before broad application in sustainable agriculture can be considered.