Comparative Modeling of Cereal Yields in Senegal: A Hybrid Approach Using Bi-LSTM and Ensemble Learning on Agro climatic FAO Data

Accurate prediction of cereal yields is critical for food 
security, particularly in Sahelian regions characterized by high 
climatic variability. This study develops a machine learning 
framework integrating dynamic agroclimatic variables (precipitation, 
temperature, soil nutrients) with FAO production statistics in Senegal 
over [2000–2024]. Feature selection based on correlation with yield 
indicated that MODIS-derived vegetation indices (NDVI, EVI, SAVI) 
were less relevant and thus excluded. Several models were evaluated, 
including Random Forest, XGBoost, CatBoost, and a Bidirectional 
LSTM explicitly designed to capture temporal dependencies. The Bi
LSTM achieved the highest predictive accuracy (R2 =0.94, 
RMSE=98.53), followed by CatBoost (R2=0.80,RMSE=216.21), 
XGBoost (R2=0.74, RMSE=243.07), and Random Forest (R2 = 0.72, 
RMSE = 251.46). Robustness was assessed using the Diebold
Mariano test, and interpretability was explored with SHAP values. 
The study demon strates that agroclimatic and production variables 
dominate over vegetation indices in predicting yields and highlights 
the trade-off between the superior accuracy of deep learning models 
and their higher computational cost. These results provide a reliable 
and interpretable framework for yield fore casting in Sahelian 
agriculture, emphasizing both methodological rigor and practical 
applicability.