Machine learning ensemble technique for exploring soil type evolution

Machine learning has shown great potential in predicting soil properties, but individual models are often prone to overfitting, limiting their generalization. Ensemble models address this challenge by combining the strengths of multiple algorithms. This study applies a voting-based ensemble model (VEM), integrating Random Forest (RF), Support Vector Machine (SVM), and XGBoost (XGB), to gain a deeper understanding of soil type evolution, which is crucial for land management and soil conservation. The research, conducted in the Tongzhou District of Beijing, uses 5,000 sampling points selected via genetic algorithms for model training, 237 surface samples for consistency testing, and 97 profiles for field validation. The VEM demonstrates high accuracy and robustness, producing a detailed soil type map and identifying key trends in soil type evolution. This study highlights the effectiveness of ensemble models in understanding soil evolution and offers valuable insights into soil system dynamics.