Improving Crop Yield Prediction Accuracy: A Hybrid Machine Learning Approach

With the world’s growing population and increase in global food demand, improving crop yield is essential to meet this rising need, reduce the impact of food production on the environment, and contribute to the United Nation Sustainable Development Goals 2 (Zero Hunger) and 13 (Climate Action). Harnessing the increasing adoption of Artificial Intelligence in diverse application areas including agriculture, this work utilises the benefits of eXtreme Gradient Boosting (XGBoost), attention mechanism and Support Vector Regression (SVR) in an ensemble for predicting agricultural yields. A remarkable R² score of 0.9863, an accuracy of 99.35%, a mean squared error of 97627518.13, and a mean absolute error of 5277.06 are the results of the rigorous evaluation of the ensemble model using 5-fold cross-validation. The cross-validation guarantees generalisability across many datasets and the ensemble’s strong performance is credited to its capacity to grasp intricate linkages in agricultural data. These results also indicate the model’s ability to outperform the existing methods including recurrent neural networks, random forest and Naive Bayes. Implications for improving agricultural resource management and decision-making may arise from this work, which represents a major step forward in crop output prediction.