Crop yield forecasting in Senegal: application of machine learning methods

Global challenges, including population growth, climate change, and food insecurity, necessitate accurate crop yield forecasting for sustainable agricultural planning. This study aimed to evaluate the effectiveness of three machine learning approaches, like Stepwise Multiple Regression, LASSO, and Random Forest, in predicting crop yields for groundnut, millet, and cotton, to identify the best-performing model suited for precision agriculture in Senegal. The prediction models were evaluated using historical agricultural and climatic data from Senegal spanning 1980 to 2021. Agricultural trend analysis reveals significant inter-annual variability by crop and period, with low variability observed from 1990 to 2000 for groundnut and millet, and from 2011 to 2021 for cotton. High variability occurred from 2000 to 2010 for groundnut and cotton, and from 1980 to 1990 for millet. Overall, area, production, and yields fluctuate significantly by period and crop. The crop yield prediction models for groundnut, millet, and cotton performed satisfactorily on the test dataset, except in some cases. The models showed strong predictive performance for groundnut and millet, with high R2 values achieved by LASSO regression (0.96 and 0.98) and Stepwise Multiple Regression (0.93 and 0.98). For cotton, both the linear regression and LASSO regression yielded an R2 of 0.70, indicating a reasonable level of accuracy. However, the Random Forest model performed poorly for cotton, with a very low R2 of 0.01. Among all models, LASSO regression consistently recorded the lowest RMSE and MAE values, making it the most effective model overall for predicting the yields of groundnut, millet, and cotton in Senegal.