Application of an Ensemble Method Based on XGBoost and Transformer in Predicting Sugarcane and Dragon Fruit Yields in Guangxi Region

This study proposes an ensemble method specifically designed for predicting sugarcane and dragon fruit yields in the Guangxi region, based on XGBoost and Transformer. After exploring the complex relationship between meteorological factors (such as temperature, humidity, and precipitation) and the yields of sugarcane and dragon fruit, we developed this innovative predictive model. By combining the strong predictive power of XGBoost with the temporal pattern recognition ability of Transformer-based models, the ensemble model demonstrated excellent accuracy, stability, and convergence during both training and testing stages. We also conducted extensive data preprocessing and feature engineering to further enhance the model's performance. Practical applications demonstrate that this method is not only applicable to the Guangxi region but also widely used in other aspects of agricultural planning, risk management, and sustainable development. Moreover, compared to a single model, the flexibility and robustness of the ensemble method have been improved. Future research directions will include the assessment of model generalization ability, the increase of interpretability, and the possibility of extending this method to more crops and regions. Overall, this study provides not only an advanced and effective method for climate-driven crop yield prediction but also robust support for cross-disciplinary research in agricultural science and economics.