Multivariate analysis and machine learning prediction of Sorghum cultivar traits under nitrogen regulation

Genotypic differences in nitrogen use efficiency strongly influence sorghum growth and yield, highlighting the need for precise and reliable prediction of cultivar responses to nitrogen (N) availability. This study investigates the impact of two N treatments on sorghum cultivars, using artificial intelligence (AI) models for prediction.

A randomized complete block design with two treatments: 0 kg N ha^- 1 (0 N) and 238 kg N ha^- 1 (238 N) was used. Six hybrid sorghum cultivars (Gustav, Estyphon, Foehn, Vegga, Aday1 and Beydarı) were evaluated for different traits. Statistical analysis included two-way ANOVA and factorial regression to assess treatment effects. Significant treatment effects were observed. Beydarı and Estyphon exhibited larger stem diameter and leaf area under 238 N, while Aday1 had the smallest values under 0 N. Gustav showed the highest panicle width, panicle weight, and grain yield under 238 N. Stomatal conductance showed an opposite trend, decreasing with N supplementation. Machine learning models, specifically Random Forest (RF) and Light Gradient-Boosting Machine (LightGBM), were used to model the interaction, achieving R² values ranging from 0.759 to 0.966 for RF and 0.729 to 0.980 for LightGBM, indicating strong predictive accuracy.

LightGBM consistently achieved R² values greater than 0.92 for key traits, such as stomatal conductance, panicle width, and grain yield, demonstrating its potential to optimize N management. Gustav performed best under high N, whereas cultivar responses to low N were genotype-specific, captured effectively by the machine learning models. These findings highlight the role of AI models in predicting cultivar performance and supporting sustainable agricultural decisions.