Prediction of construction and demolition waste using remote sensing-GIS integration and stacking ensemble learning

A waste management strategy requires accurate estimation of construction and demolition waste (CDW) generation rates. Most existing studies predominantly focus on project-level estimation, rely on publicly available data, employ standalone machine learning (ML) models, and provide limited interpretability. This study aims to develop a robust framework for predicting CDW quantities based on demographic, economic, environmental, and construction-related drivers. The main contribution lies in integrating remote sensing and geographic information system techniques to construct a comprehensive geospatial database, which was subsequently utilized within ML models to simulate and forecast regional-level CDW generation. Six state-of-the-art ML models, including extreme gradient boosting (XGB) and random forest (RF) were applied and their performance was assessed using four evaluation metrics. Among these, XGB and RF emerged as the top-performing models, and their integration within a stacking ensemble further improved prediction accuracy (R2 = 0.93, EVS = 0.93, RMSE = 460,880.96, and MAE = 331,259.05). SHapley Additive exPlanations analysis showed that the numbers of residential units and buildings were the most important features influencing CDW generation. The findings support applications in urban metabolism monitoring, planning of waste management infrastructure, and advancement of circular economy strategies within the study region.