Dynamic Cotton Yield Prediction Using Machine Learning and Statistical Modeling: Application to the Climatic Conditions of Chad

Crop yield forecasting represents a major strategic challenge for developing countries,where agriculture remains highly dependent on climatic conditions [3,24]. In Chad,cotton cultivation constitutes a key pillar of the national economy; however, itsproductivity is strongly affected by interannual climate variability [25,26]. The absenceof reliable yield forecasting systems limits decision-making by farmers, financialinstitutions, and public policymakers [23].This paper proposes a comprehensive methodological framework for the dynamicprediction of cotton crop yield, based on supervised statistical models related tomachine learning, applied to long-term climatic time series covering the period 1985–2025. A comparative analysis is conducted between Multiple Linear Regression (MLR),Principal Component Regression (PCR), and Partial Least Squares Regression (PLS)in the Logone Occidental region of Chad.Model performance is evaluated using standard statistical indicators (RMSE, MAE,nRMSE, coefficient of determination R², and correlation coefficient r), in accordancewith methodological recommendations established in the literature [27,28]. The resultsshow that PLS consistently outperforms the other models, particularly in contextscharacterized by strong collinearity among climatic variables and over long time series.The optimal model is integrated into a decision support system, highlighting thepotential of data-driven approaches for sustainable and climate-resilient agriculture inSub-Saharan Africa.