Compressive Strength Prediction of Fly Ash Geopolymer Concrete Using Support Vector and Random Forest Regression

Abstract

The mix design of geopolymer concrete includes consideration of several factors and many trail tests. Obviously, to obtain specified target strength different combinations need to be repeatedly adjusted. As a result, this process consumes more time and energy resources. Therefore, from the sustainability of resources it is essential to develop a predictive model, which can show the influence of different combinations on the targeted compressive strength of geopolymer concrete. The current work includes prediction of compressive strength of fly ash based geopolymer concrete using machine leaning algorithms. The input data sets for the modelling are collected from previously conducted research works. Machine leaning algorithms employed are Support Vector Regression (SVR) and Random Forest method (RFM). The input variables include quantity of sodium silicate solution, molarity of sodium hydroxide, ratios of NaOH to Na2SiO3, Alkaline liquid to fly ash and total water content to geopolymer solids. About 75 % of the data collected is used for training and 25 % of the data is used for testing the performance of the model. The accuracy of the model developed is checked using coefficient of determination. It is observed from statical checks that both methods predicted the compressive strength to an acceptable degree of accuracy. The coefficient of determination of for SVM and RFM are found to be 0.82 and 0.81 respectively. Results indicated that the quantities of aggregates are not influencing the compressive strength of geopolymer concrete, on the other hand alkaline liquid to binder and water to geopolymer solids are showing significant impact on the strength of mixes.