Machine learning in the construction industry: potential of artificial neural networks in estimating construction and demolition waste

The estimation of construction and demolition waste (CDW) generation is essential for sustainable planning and effective waste management on construction sites. However, conventional methods often fail to meet the practical demands of the sector. This study investigated the use of artificial neural networks (ANN) as a predictive tool for CDW quantification. Simulations were performed with samples of 5,000 data points (A) and 10,000 data points (B), followed by validation with a sample of 360 data points (R) collected from construction sites in Curitiba, Paraná state, Brazil. This approach allowed for a comprehensive evaluation of the predictive accuracy and practical applicability of the ANN. The best performance was obtained with sample B, using an ANN configured with two input variables, ten neurons in the hidden layer, and three training cycles. In the simulations, the model presented a coefficient of determination (R²) of 1.00, a root mean squared error (RMSE) of 6.55 kg, and a mean absolute percentage error (MAPE) of 0.00013%. In the validation, an R² of 0.83 was obtained, along with an RMSE of 4,337.69 m³, and accurate estimates in over 60% of cases (MAPE). The results demonstrated the viability of using ANNs to improve CDW estimation, contributing to decision-making and the development of more efficient waste reduction strategies in civil construction.