Utilization of support vector-based techniques for prediction of compressive strength for graphene oxide based self- compacting concrete

In the present work, the machine learning methods were used to develop the predictive models for the Compressive Strength (CS) of the graphene oxide-enhanced self-compacting concrete. The models were built and analysed using a comprehensive database of 76 samples extracted from the literature. The model uses water content, ground granulated blast furnace slag, fly ash, fine aggregates, coarse aggregates, superplasticizer and cement quantity as input parameters and concrete CS as output. The predictive modelling methods used are linear regression, genetic algorithm, particle swarm optimization, support vector regression, grey wolf optimization, differential evolution and manta rays foraging optimization. The model efficiency was assessed on the basis of seven parameters, namely, coefficient of correlation (CC), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), normalized root mean square error, Willmott’s index, Nash Sutcliffe efficiency and coefficient of determination. The results showed that the support vector regression differential equation had the highest accuracy and lowest error to predict CS. The maximum CC obtained were 0.990 and 0.960 for training and testing datasets respectively. The error values for training and testing datasets are SVR DE achieved CC values of 0.995 and 0.958, RMSE values of 1.785 and 3.947 MPa, MAE values of 0.874 and 2.899 MPa, NRMSE values of 0.023 and 0.088, WI values of 0.997 and 0.978, and NSE values of 0.990 and 0.911 respectively.