• • The random forest model predicted that 40.91% of farmland is suitable for Se-rich maize, a 25.86% increase over soil total Se-based assessment, enabling more efficient land use in Se-poor regions.
• • 53.64% of maize grain samples met the Se-rich product standard (0.02–0.30 mg·kg−1) despite low soil Se, demonstrating that soil total Se alone is insufficient for evaluating crop Se status.
• • Soil pH, CaO, and MgO were significantly positively correlated with the Se bioaccumulation coefficient, serving as effective proxies for soil available Se in predictive modeling.
• • The RF model outperformed multiple linear regression in predicting maize Se content, offering higher accuracy and reliability for agricultural planning in Se-deficient areas.