Architecture of Arcitficial Neural Network in Prediction of Sustainable Concrete Compressive and Split Tensile Strength

Abstract

Artificial neural networks are utilized in many fields as well as in civil engineering applications. One of these applications is compressive and split tensile strength prediction. Number of layers in neural network and number of neurons in each hidden layer are determinant factor of ANN model performance. In general practice, number of hidden layers are selected first then number of neurons in each hidden layer is determined by considering the complexity of the relationship between input and output of parameters. Yet, there is no accepted practice or set of rules in the literature. The goal of this research is to investigate effect of number of neurons in ANN architecture in sustainable concrete compressive and split tensile strength prediction. Total of 2551 iterations were performed, and 144 number of different ANN architectures were tested. In this research best coefficient of correlation (R2) value was determined to be 0.98419 in the ANN architecture where first hidden layer contains 5 and second hidden layer contains 13 neurons. The data set utilized in ANN consists of 321 number of test results with 8 inputs and 2 outputs. In ANN architecture the inputs are water, cement fine aggregate, recycled aggregate, natural coarse aggregate, superplasticizer, density, absorption, and outputs are; compressive strength (CS) and split tensile strength (STS).