COMPARISON OF STATISTICAL AND SOFT COMPUTING METHODS FOR ESTIMATION OF COMPRESSIVE STRENGTH OF FIBER REINFORCED CONCRETES

Abstract

This study presents soft computing and statistical models for the prediction of the compressive strength of steel fiber-reinforced concrete. The model examines nine different parameters for the matrix mixture, including: (C) cement content; (W) water content; (CA) coarse aggregate content; (FA) fine aggregate content; (Dmax) maximum size of aggregate; (Vf) volume of steel fiber; (Lf) length of fiber; (Df) diameter of fiber; and (SP) superplasticizer content. An artificial neural network (ANN), gene expression programming (GEP), random forest method (RFM), LASSO regression, and multiple linear regression (MLR) methods were used to predict the 28-day compressive strength. The data set was formed by collecting 204 experimental data samples from available literature to cover different variables. These models are compared and evaluated through statistical tests such as coefficient of determination (R2 ) and root mean squared error (RMSE). Statistical analysis and comparisons revealed that ANN, GEP, and RFM models can be useful tools in predicting the compressive strength of steel fiber-reinforced concrete. The results also indicate that the proposed formula for ANN, GEP, and RFM provides a more reliable and accurate prediction ability than the LASSO and MLR methods.