Advanced Efficient Feature Selection Integrating Augmented Extreme Learning Machine and Particle Swarm Optimization for Predicting Nitrogen Use Efficiency and Yield in Corn

Josselin Bontemps, Isa Ebtehaj, Gabriel Deslauriers, Alain N. Rousseau, Hossein Bonakdari, Jacynthe Dessureault-Rompré

Efficient nitrogen management is crucial for improving corn productivity while minimizing environmental impacts. This study evaluates the response of corn to nitrogen fertilization using three key metrics: yield; nitrogen harvest index (NHI); and agronomic nitrogen use efficiency (ANUE). This experiment was conducted over three years (2021–2023) across 84 sites in Quebec, Canada, with five nitrogen treatments applied post-emergence (0, 50, 100, 150, 200 kg N/ha) and initial nitrogen applied at seeding (30 to 60 kg/ha). In addition, various soil health indicators, including physical, chemical, and biochemical properties, were monitored to understand their interaction with nitrogen use efficiency. Machine learning techniques, such as augmented extreme learning machine (AELM) and particle swarm optimization (PSO), were employed to optimize nitrogen recommendations by identifying the most relevant features for predicting yield and nitrogen use efficiency (NUE). The results highlight that integrating soil health indicators such as enzyme activities (β-glucosidase [BG] and N-acetyl-β-D-glucosaminidase [NAG]) and soil proteins into nitrogen management models improves prediction accuracy, leading to enhanced productivity and environmental sustainability. These findings suggest that advanced data-driven approaches can significantly contribute to more precise and sustainable nitrogen fertilization strategies.