Towards sustainable construction: estimating compressive strength of waste foundry sand-blended green concrete using a hybrid machine learning approach

Concrete production necessitates a large quantity of fine aggregate. The use of waste foundry sand (WFS) to replace natural fine aggregate significantly enhances the sustainability of the construction industry. This study proposes and verifies a novel approach for estimating the compressive strength (CS) of concrete mixes that contain WFS. The research methodology involves a literature review to identify explanatory variables for CS estimation and the construction of a historical dataset, which consists of 430 samples. This dataset is used to train and verify a hybrid machine learning method, which is an integration of Light Gradient Boosting Machine (LightGBM) and Biogeography-Based Optimization (BBO). LightGBM is used to construct a functional mapping between the CS and its influencing factors. BBO optimizes the training phase of the machine learning approach. The proposed model is evaluated using root mean square error (RMSE), mean absolute percentage error (MAPE), and coefficient of determination (R2). Notably, an asymmetric loss function is employed during the training phase of LightGBM to restrict overestimations. Finally, a Python program with a graphical user interface is developed to facilitate practical application of the new approach. Experimental results show that the proposed model, named BBO-LightGBM, has achieved good predictive performance with an RMSE of 3.92 and a MAPE of 8.46%. BBO-LightGBM is capable of explaining roughly 95% of the variability in the response. Moreover, the proposed framework is able to reduce the proportion of overestimations by roughly 18%. This study demonstrates the potential of BBO-LightGBM in providing accurate and reliable estimations of the CS of WFS-blended concrete.