In the face of climate change, soil degradation, and rising food demand, optimizing corn production with sustainable inputs like arbuscular mycorrhizal fungi (AMF) and plant growth-promoting rhizobacteria (PGPR) is essential. This study integrates 73 plant-soil features—including 41 engineered interaction terms—with advanced machine learning to predict grain yield more accurately than traditional methods.

We compared eight algorithms and found that ANFIS (R² = 0.555), Transformer (R² = 0.545), and ANN (R² = 0.518) outperformed others, highlighting the power of neural networks and attention mechanisms in capturing nonlinear ecophysiological relationships. Key drivers include canopy temperature during critical stages, interactions like Leaf Area Index × Dry Matter Yield, and root colonization effects—revealing how biofertilizers enhance nutrient uptake, stress tolerance, and yield formation.

Using SHAP interpretations, partial dependence plots, and feature clustering, we uncover hidden pathways linking microbial symbiosis, photosynthesis efficiency, and final yield. This work advances precision agriculture, supports climate-resilient practices, and promotes input-efficient systems toward sustainable food security.

The full article is available here: https://doi.org/10.1038/s41598-026-40919-3