DIGREP: a grey-relational deep ensemble model for construction waste prediction

Accurate forecasting of construction waste is critical for sustainable resource management and environmental protection. This study presents a deep integrated grey-relational ensemble predictor model that combines grey relational analysis for feature selection with a stacking ensemble of backpropagation network, convolutional neural network and long short-term memory neural networks, with random forest as the meta-learner. The model integrates generation source attributes as well as material characteristics of construction waste. Experimental results on datasets from Eurostat and the U.S. Environmental Protection Agency show that the model achieves a mean absolute error of 5.4 tonnes, a mean squared error below 12.6 tonnes² and an R² of 0.913. In addition, it maintains a waste processing response time within 0.8 s and realises a recovery rate of 89.7%. The model effectively addresses multi-source heterogeneity and complex non-linear interactions, offering a robust solution for both short- and long-term waste prediction. These findings support smart waste planning and offer practical value for the circular economy and low-carbon construction.