Modeling Compressive Strength of Lightweight Geopolymer Mortars by Step-Wise Regression and Gene Expression Programming

Abstract

This article presents a comprehensive study aimed at developing suitable mathematicalmodels for the prediction of compressive strength of lightweight geopolymer mortar(LWGM) with different types and amounts binders with different curing regimes. Lightweightpumice aggregate, alkali activated powder materials are the main components of geopolymerbinder. From the experimental study 306 data samples were obtained and these wereused to derive explicit formulas for estimation of the compressive strength of LWGMs. Twomethods are used to produce the models. The first is the simplified linear step-wise regression,while the second method is the genetic expression programming. Step-wise regressionis a statistical tool that uses the impact of each factor to evaluate its effect on the equation.This impact is calculated based on the probability effect based on the F-distribution and thenull-hypothesis. The default value of probability that refers to the significance of each factoris 0.05. Thus, the software calculates the probability of each of the independent variables andincludes only those with probability values less than 0.05. Based on the included independentvariables, simplified linear regression equation is introduced. The genetic programming onthe other hand, is much more sophisticated method that uses the principles of gene evolution.The modeling is separated for each type of binder. Thus, two sets of formulas are obtainedfrom each modeling, one for the granulated blast furnace slag -based LWGM, while thesecond is for the fly ash-based LWGM. These models revealed that genetic algorithm basedmodeling has a reliable potential for estimating the strength of LWGMs.