Enhancement of Sorghum Forecasting Models Using Machine Learning in the rain-fed sector in Sudan

Sorghum, a major global crop, is prominently produced in Sudan, ranking among the top sorghum producers globally. This production is concentrated in El-Gadarif State, covering about 78,228 square kilometers, and relies heavily on both irrigated and rain-fed cultivation, particularly of dry sorghum. The unpredictable and sparse nature of rainfall in this region underscores the critical need for accurate forecasting to optimize crop yield. Addressing this, the study introduces a machine learning-based prediction system that utilizes historical rainfall and yield data from 1943 to 2021. The research focuses on developing and evaluating neural network architectures, employing multi-layer perceptrons (MLP) and XGBoost regressor models. Notably, the XGBoost regressor outperforms the MLP regressor, displaying a significantly higher accuracy with an R
²
value of 0.87 compared to 0.25. This highlights the crucial role of meticulous model calibration and selection. Importantly, the superiority of the XGBoost regressor becomes particularly evident in scenarios with limited data and is further emphasized after extensive calibration efforts.