Prediction of Waste-to-Energy Potential from Municipal Solid Waste Using Random Forest Regression

Municipal solid waste (MSW) is increasingly considered a feedstock for waste-to-energy (WtE) systems, yet recoverable electricity depends strongly on waste composition and moisture. This study applied Random Forest regression to predict net electricity yield from MSW and identify its main drivers. A synthetic dataset of 600 MSW scenarios was generated from literature-informed ranges for waste generation, organic, paper, plastic, other combustible and inert fractions, and moisture content. Electricity yield was calculated from a lower heating value-based energy balance. The data were divided 80:20 for training and testing, and Random Forest hyperparameters were tuned using five-fold cross-validation. Multiple linear regression and a decision tree were used as baseline models. Random Forest achieved a test R² of 0.835, RMSE of 30.1 kWh/t, and MAE of 24.6 kWh/t. Moisture content and plastic fraction were the strongest predictors, with importance values of 54.5% and 20.7%, respectively. Predicted yield declined from 474.1 to 398.3 kWh/t across low to high organic-fraction classes. The results indicate that moisture reduction is a key technical lever for improving WtE potential, although validation with measured Indonesian waste data remains necessary.