Artificial neural network and soft computing models to predict the compressive strength in self-compacting green concrete

This study explores self-compacting concrete (SCC) enhancement by incorporating silica fume, fly ash, or both across various mix compositions. The research evaluates six predictive models—linear regression, nonlinear regression, pure quadratic, interaction, full quadratic, and artificial neural network (ANN)—to predict the compressive strength of SCC. A dataset of 330 experimental studies covering a wide range of parameters, such as water/cement ratio, cement content, aggregate content, superplasticizer content, silica fume content, fly ash content, and curing time, is used. The compressive strength of the datasets ranges from 4.9 to 87 MPa, while slump flow diameter ranges from 450 to 790 mm. The models are assessed using objective function, root mean square error (RMSE), scatter index (SI), and mean absolute error (MAE). The ANN is the most accurate among the models, achieving an R² of 0.94, RMSE of 3.56 MPa, MAE of 2.67 MPa, and SI of 0.09. The models effectively predict compressive strength across various concrete compositions, although they do not predict slump flow diameter, as SCC specifications require it to be within 550 to 850 mm.