Prediction of concrete compressive strength with GGBFS and fly ash using multilayer perceptron algorithm, random forest regression and k -nearest neighbor regression

In this study, supervised learning and neural networks were applied to predict the compressive strength of concrete mixes with GGBFS and fly ash. Three models: Multilayer perceptron network (MLP), random forest regression (RFR) and k -nearest neighbor (KNN) regression methods were employed using Python to estimate the compressive strength of concrete mixes. Inputs included cement content, water content, coarse aggregate, fine aggregate, superplasticizer and maturity age, and output was concrete compressive strength. The three methods were compared according to their accuracy and stability to predict compressive strength. Results showed that RFR and MLP regression produced close results and both had better performance and produced less amount of error compared to KNN. Stability results showed that RFR was the least influenced by the data splitting process and it was addressed as the most stable model.