Estimation of strength, rheological parameters, and impact of raw constituents of alkali-activated mortar using machine learning and SHapely Additive exPlanations (SHAP)

One-part alkali-activated material (AAM) is a new eco-friendly developed low-carbon binder that utilizes alkaline activators in solid form. This study deals with the experimental synthesis of one-part alkali-activated mortar (AAM) based on the partial replacement of fly ash (FA) with hydraulic lime (LM) as a precursor, and machine learning-based gene expression modeling (GEP) modeling for the optimization of the developed AAM. The datasets were established by the experimental work performed during this current study. The chosen input parameters were fly ash, hydraulic lime, sodium silicate, sodium hydroxide, sand/binder, water/binder, curing age, and time after mixing. The experimental results showed greater compressive strength and rheological pa-rameters for the specimens having a high quantity of hydraulic lime. The GEP model has shown a strong generalization capability and prediction capacity for the future estimation of compressive strength, plastic vis-cosity, and yield strength. All the models showed a strong correlation of 0.92, 0.89, and 0.96 for compressive strength, plastic viscosity, and yield stress respectively. SHapely Additive exPlanations (SHAP) were employed to explore the effect of each input parameter of AAM on the predicted outcomes. The results revealed a strong interaction and positive effect of LM on the YS and PV while a negative impact was observed for the compressive strength. While fly ash has shown a negative impact on three outcomes of PV, YS, and CS respectively. The addition of LM and SS-activator leads to earlier structural build-up due to the flocculation of particles caused by the faster geopolymerization reactions.