New random intelligent chemometric techniques for sustainable geopolymer concrete: low-energy and carbon-footprint initiatives

The construction industry, being a significant contributor to greenhouse gas emissions, faces considerable attention and demand on account of the increasing global apprehension regarding climate change and its adverse impacts on the environment. Geopolymer shows itself as a viable and sustainable alternative to the Portland cement binder in civil infrastructure applications, offering a low-energy, low-carbon-footprint solution. This study evaluates five models: random forest (RF), robust linear regression (RL), recurrent neural network (RNN), response surface methodology (RSM), and regression tree (RT). The RL and RT models were utilized in the prediction of GPC compressive strength (CS), employing the Matlab R19a regression learner APP. The RNN model was implemented using the Matlab R19a toolkit. Furthermore, the RF model was developed using R studio version 4.2.2 programming code, and the RSM model was constructed using the Minitab 18 toolbox. EViews 12 software was utilized for both pre-processing and post-processing of the data. Additionally, it was employed to convert the non-stationary data into stationary data to obtain accurate results. The input variables included SiO₂/Na₂O (S/N), Na₂O (N), water/binder ratio (W/B), curing time (CT), ultrasonic pulse velocity (UPV), and 28-day compressive strength (MPa) (CS) as the target variable. The findings of the study indicate that the RMS-M3 model exhibited superior performance compared to all other models, demonstrating a high level of accuracy. Specifically, the Pearson correlation coefficient (PCC) was calculated to be 0.994, while the mean absolute percentage error (MAPE) was found to be 0.708 during the verification phase.