Hybrid Machine Learning and Metaheuristic Optimization for Hickory Yield Prediction: An Empirical Evaluation of SVR, XGBoost, and RF with AEO and ArchOA

Abstract

Accurate crop yield predictions are vital for sustainable agriculture and resource efficiency, benefiting farmers, agronomists, traders, and policymakers by informing decisions on planting, harvesting, management, trading strategies, and policy. This study explores advanced machine learning methods RF, SVR, and XGBoost for forecasting hickory yields, enhanced by hyperparameter optimization using AEO and ArchOA. Developing and evaluating four predictive schemes, the study employs an 80% training and 20% testing data split for robustness and accuracy. The dataset consists of 52 samples, with variables such as temperature, rainfall, and other related factors. Initial RF, SVR, and XGBoost evaluations are followed by hybrid schemes integrating their strengths for improved accuracy, further refined through systematic hyperparameter optimization. Evaluation metrics include RMSE (Root Mean Squared Error) and R² (Coefficient of Determination). In the training phase, XGBoost-AEO had the best performance with an RMSE value of 69.61354 and R² of 0.999619. In the testing phase, the XGBoost-AEO model outperformed other models with RMSE of 742.607 and R² of 0.959451. Results demonstrate the superior performance of hybrid schemes, especially SVR-AEO and XGBoost-AEO, highlighting the effectiveness of advanced machine learning and optimization techniques in enhancing crop yield predictions and supporting sustainability and food security objectives.