Ensemble Methods for Time Series Forecasting in Nigeria: Predicting Agricultural Yields Using Advanced Machine Learning Approaches

International audience Accurate forecasting of agricultural yields is essential for maintaining food security, economic stability, and sustainable resource management in Nigeria. This research seeks to enhance the accuracy and reliability of yield predictions for three staple crops—maize, rice, and cassava—by employing ensemble methods. The study examined a comprehensive dataset covering 24 years (2000-2023) and investigated various individual forecasting models, including linear regression, ARIMA, Random Forest, Gradient Boosting, Support Vector Machines, and Neural Networks. Three ensemble methods—bagging, boosting, and stacking—were utilized, and the performance of the models was assessed using metrics such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE), along with statistical significance tests using R package. The findings indicated that ensemble methods significantly surpassed individual models across all metrics and crop types. The stacking ensemble method achieved the highest accuracy, reducing RMSE by 23.7% compared to the best individual model. This study offers compelling evidence that ensemble methods can substantially improve the accuracy of agricultural yield forecasting in Nigeria. These results have significant implications for food security planning, agricultural policy formulation, and resource allocation strategies, contributing to a more profound understanding of precision agriculture.