Foretelling the compressive strength of concrete using twin support vector regression

Characteristic compressive strength is a key and crucial physical attribute of concrete used in various design standards and rules. In this study, twin support vector regression (TSVR) is employed to foretell the concrete compressive strength (CCS) of high strength concrete (HSC). TSVR is a relatively new method that has shown strong generalization performance and rapid learning speed in many regression applications. The TSVR algorithm was constructed using datasets from the existing literature. Its outputs were then compared with those of preexisting models employed to foretell the strength of HSC. A total of 324 datasets from previous studies were used to train the models. The input variables employed for predicting CCS include superplasticizer (S_U), fine aggregate (FA), coarse aggregate (CA), water (W), and cement (C). TSVR demonstrates commendable performance when compared to various other models, including artificial neural network (ANN), bagging regression trees (BRT), fuzzy polynomial neural networks (FPNN), genetic operation trees (GOT), neural-fuzzy inference system (NFIS), support vector machine (SVM) and others. Performance parameters, namely the coefficient of determination (R²), coefficient of correlation (R), mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE), indicate that the TSVR algorithm is highly capable of accurately foretelling the CCS.