Using tree-based machine learning models to predict diverse compost maturity via one-hot encoding: Model deployment,...

This study innovatively integrates various classification features like initial properties of composting materials and scale, with process parameters like time, temperature, pH, etc., to predict compost maturity of various types using the seed germination index (GI) with tree-based machine learning (Random Forest, Extra-Trees, Gradient Boost, AdaBoost, XGBoost, and LightGBM). The results indicate that the AdaBoost model achieves a high prediction accuracy (R² = 0.9720) with low errors (RMSE = 5.3495, MAE = 2.7872), outperforming other models. Feature importance on the Gini index and SHAP analysis reveals that in process parameters, composting time and pH are vital factors influencing maturity, while the type of compost material and seed variety also significantly impact GI. A stacking method combined the mentioned models to enhance prediction accuracy and ubiquity, resulting in an improved R² of 0.9733. Various animal and plant-based organic wastes were used to validate the fusion model's effectiveness, which resulted that accurately predicted maturation juncture, especially for high-nitrogen source materials. Based on these findings, an easy-to-use online application was developed, providing a practical tool for predicting compost maturity and promoting the safe application of organic fertilizers in agriculture. This application aims to assist farmers and agricultural professionals in optimizing compost usage, enhancing soil quality, and ensuring sustainable farming practices.