Predicting concrete’s compressive strength (CS) is a crucial and challenging task in civil engineering as it directly impacts the longevity and structural integrity of infrastructure initiatives. Precise estimation of the water–cement ratio (W/C) is essential for guaranteeing the structural integrity of structures since it is a critical parameter that greatly affects concrete’s CS. This study carries out an extensive investigation of the prediction of the W/C of concrete, utilizing the enormous potential of machine learning, including the backpropagation neural network (BPNN), bilayer neural network, boosted tree algorithm, bagged tree algorithm (BGTA), and support vector regression (SVR), using 108 datasets. We integrate artificial intelligence models with traditional engineering techniques to develop a reliable, precise, and efficient forecasting system. The study input includes curing days (D), fiber (F), cement (C), fine and coarse aggregate (FA and CA), density (Den), CS, water (W), and W/C as the output variables. The result shows that, in comparison to the other models, BGTA-M3 achieved the best performance evaluation criterion. In the calibration and verification phases, NSE, PCC, R, and WI = 1 and MAPE = 0.00, respectively. BPNN-M3 had an MAPE of 0.0004 in the verification phase. The study uses SHapley Additive exPlanations (SHAP), an explainable artificial intelligence (AI) technique, to improve decision-making in complex systems, with cement “C” significantly contributing to higher predictions in SVR-M2. Future studies should expand the dataset to include information from diverse geographic areas, environmental conditions, and concrete mixes to enhance the applicability and dependability of the models.
Read full abstract