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

Area of Science:

• Agricultural Science
• Environmental Science
• Data Science

Background:

• Compost maturity is crucial for effective soil amendment and nutrient management.
• Predicting compost maturity traditionally relies on time-consuming laboratory analyses.
• Accurate maturity prediction is essential for safe application of organic fertilizers.

Purpose of the Study:

• To develop and validate machine learning models for predicting compost maturity.
• To identify key features influencing compost maturity prediction.
• To create a practical tool for agricultural professionals to assess compost quality.

Main Methods:

• Integration of composting material properties and process parameters (time, temperature, pH).
• Application of tree-based machine learning algorithms: Random Forest, Extra-Trees, Gradient Boost, AdaBoost, XGBoost, and LightGBM.
• Utilizing seed germination index (GI) as the primary maturity indicator.
• Employing feature importance analysis (Gini index, SHAP) and a stacking ensemble method.

Main Results:

• The AdaBoost model demonstrated high prediction accuracy (R² = 0.9720) and low errors (RMSE = 5.3495, MAE = 2.7872).
• Composting time and pH were identified as critical process parameters influencing maturity.
• A stacking model achieved an enhanced accuracy of R² = 0.9733.
• The developed fusion model accurately predicted maturation for various organic wastes, especially high-nitrogen materials.

Conclusions:

• Machine learning, particularly ensemble methods, offers a robust approach to predicting compost maturity.
• Composting time and pH are key indicators for assessing compost quality.
• An online application was developed, providing a user-friendly tool for practical compost maturity assessment.
• The findings support the safe and optimized application of organic fertilizers, promoting sustainable agriculture.