Prediction of compressive strength and ultrasonic pulse velocity of fiber reinforced concrete incorporating nano silica using heuristic regression methods

In this study, five different Artificial Intelligence (AI) models, Multivariate Adaptive Regression Splines (MARS), M5P Model Tree (M5P-MT), Least Square Support Vector Machines (LS-SVM), Multilayer Perceptron Neural Network (MLP-NN) and Multiple Linear Regression (MLR) have been developed to predict the Compressive Strength (CS) and Ultrasonic Pulse Velocity (UPV) of Fiber Reinforced Concrete (FRC) incorporating nano silica. Experimental results from 175 and 132 concrete samples with different mixture proportions were collated, respectively, to develop the models for CS and UPV. Standard statistical performance evaluation measures such as the Correlation Coefficient (CC), Root Mean Square Error (RMSE), Mean Absolute Percentage of Error (MAPE), Performance Index (PI), Average Absolute Error (AAE), Standard Deviation (SD) and mean (M) were used to evaluate proposed models in training and testing stages. The MARS model using normalized input data performed better compared to LS-SVM, M5P-MT, MLP-NN and MLR in the prediction of both UPV and CS. The robustness of the developed AI predictive models was verified through external validations and the results indicate that proposed models are robust and they provide accurate predictions of CS and UPV.