Estimation of Compressive Strength of Rubberised Slag Based Geopolymer Concrete Using Various Machine Learning Techniques Based Models

Abstract

In the quest for sustainable construction practices, researchers have been exploring alternative materials that can reduce the reliance on traditional cement in concrete production. Geopolymer concrete (GPC) has surfaced as a promising alternative due to its potential ecological benefits. The formulation of GPC mixtures is a challenging task as there is no specific code provision to determine the mix design. The complexity of determining the optimal mix proportions is compounded by the influence of various factors, including the Na₂SiO₃/NaOH ratio, the quantities of sodium silicate (Na₂SiO₃) and sodium hydroxide (NaOH), and differing curing periods, all of which significantly impact the concrete’s mechanical properties. A variety of predictive modeling techniques, including multivariate adaptive regression splines (MARS), group method of data handling (GMDH), M5P, and linear regression (LR), are used in the estimation of the compressive strength of rubberized slag-based GPC. This study utilizes a dataset comprising 186 observations, which are divided into a training dataset of 130 observations and a testing dataset of 56 observations. The investigation considers various input parameters such as the molarity of NaOH (S), Na₂SiO₃ quantity (SS), sand quantity (S), coarse aggregate quantity (CA), NaOH quantity (M), the quantity of copper slag (C), rubber aggregate quantity (RA), curing period (D), and fly ash (FA), with the compressive strength serving as the output constraint. The efficacy of these approaches is assessed using performance indices such as the coefficient of correlation (CC), Nash–Sutcliffe efficiency (NSE), mean absolute error (MAE), root mean square error (RMSE), and scattering index (SI). The findings indicate that the MARS model outperforms the other soft computing techniques, with a testing CC of 0.9634, MAE of 1.4509, RMSE of 1.8465, SI of 0.0480, and NSE of 0.9265. Conversely, the LR model exhibits the least favourable performance, with testing values of CC at 0.8640, MAE at 3.0411, RMSE at 3.5375, SI at 0.0920, and NSE at 0.7303. These results emphasize the potential of MARS as a suitable method for predicting the compressive strength of rubberized slag-based GPC, leading to more sustainable construction methodologies.