Multi-model machine learning framework for prediction of greenhouse gas emissions during composting of organic solid waste.

Accurate prediction and regulation of greenhouse gas (GHG) emissions during composting are essential for mitigating global warming. Herein, two multi-model machine learning frameworks were developed based on Gradient Boosting Regression, Extreme Gradient Boosting, and Extra Trees Regression algorithms. In particular, the modular multi-model demonstrated superior performance in predicting CO2, CH4, and N2O emissions, achieving R2 values of 0.9662, 0.9729, and 0.9051, respectively. The stacking ensemble model also exhibited high accuracy in cumulative emission prediction (R2 = 0.9278). SHapley Additive exPlanations analysis identified the time of composting cycle length as the dominant factor for CO2 and total GHG emissions, while aeration rate and moisture content were the key drivers of CH4 and N2O emissions, respectively. Experimental validation confirmed accurate prediction of overall cumulative GHG emissions across the full composting cycle, with prediction accuracy above 90 % in the late stage. This work provided valuable guidance for GHG mitigation during organic waste management.