Integrating numerical methods and machine learning to optimize agricultural land use

Abstract

In the current context, optimizing the utilization of agricultural land resources is increasingly vital for production intensification. This study presents a methodological approach employing numerical methods and machine learning algorithms to analyze and forecast land use optimality. The objective is to develop effective models and tools facilitating rational and sustainable agricultural land resource management, ultimately enhancing productivity and economic efficiency. The research employs data dimensionality reduction techniques such as principal component analysis and factor analysis (FA) to extract key factors from multidimensional land data. The simplex method is utilized to optimize resource allocation among crops while considering constraints. Machine learning algorithms including extreme gradient boosting (XGBoost), support vector machine (SVM), and light gradient boosting machine (LightGBM) are employed to predict optimal land use and yield with high accuracy and efficiency. Analysis reveals significant differences in model performance, with LightGBM achieving the highest accuracy of 99.98%, followed by XGBoost at 95.99%, and SVM at 43.65%. These findings underscore the importance of selecting appropriate algorithms for agronomic data tasks. The study's outcomes offer valuable insights for formulating agricultural practice recommendations and land management strategies, integrable into decision support systems for the agricultural sector, thereby enhancing productivity and production efficiency.