Application of LightGBM for Agricultural Yield Forecasting from Environmental and Pesticide Inputs

Abstract

Accurate agricultural yield forecasting plays a critical role in supporting food security, resource management, and sustainable farming practices. The increasing availability of environmental and agricultural management data has enabled the adoption of machine learning approaches to improve prediction reliability. However, conventional statistical and learning models often struggle to capture the complex nonlinear relationships between climatic factors and pesticide application patterns. This study proposes the application of the Light Gradient Boosting Machine (LightGBM) for agricultural yield forecasting using integrated environmental and pesticide input features.The main motivation of this research is to develop an efficient and accurate forecasting framework capable of handling high-dimensional agricultural datasets while maintaining strong generalization capability. The proposed model combines environmental variables with pesticide-related indicators to provide a comprehensive representation of crop growth conditions. LightGBM is employed due to its histogram-based learning strategy and leaf-wise tree growth mechanism, which enhance predictive accuracy and computational efficiency.Model performance was evaluated using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE), along with visual comparison between actual and predicted yield values. Experimental results demonstrate that the proposed approach achieves reliable forecasting performance, with MAE of 63,522.74 kg/ha and RMSE of 82,716.32 kg/ha, indicating effective modeling of nonlinear agricultural dynamics. Future work will focus on integrating remote sensing data, advanced temporal modeling techniques, and explainable artificial intelligence to further enhance prediction accuracy and interpretability.