Predicting the compressive strength of concrete using neural network and kernel ridge regression

Abstract

Kernel methods and neural networks (NN) are two of the most powerful tools of machine learning to solve the engineering and science problems. In this paper, we propose kernel ridge regression (KRR) and NN to estimate the compressive strength (CS) of concrete with recycled aggregate based on the values of cement, natural aggregate, recycled aggregate, sand, and water. We collected a dataset of 182 samples and carried out the data analysis for each material used in making concrete. The dataset is used for training, validation, and testing of concrete's CS using KRR and NN. We use both linear and nonlinear kernels in ridge regression, and up to four layers NN with three hidden layers and one output layer for CS prediction. KRR and NN are designed in MATLAB, and we use different NN types and training functions with objective evaluation criterion to obtain the optimum models for CS prediction. We choose gradient descent with momentum weight bias learning and mean square error (MSE) performance functions for model training. Simulation results show that both KRR and NN predict the CS of concrete accurately and efficiently with admissible MSE and correlation coefficient close to one. KRR with nonlinear 4th order polynomial kernel outperforms the NN for CS prediction.