Uncertainty quantification and sensitivity analysis of geopolymer mortar compressive strength using statistical and machine learning methods

Abstract

This study investigates the uncertainty in the compressive strength of geopolymer mortars through statistical analysis from experimentally obtained data and a sensitivity analysis to quantify the variability. Geopolymer mortars are prepared with fly ash and GGBFS as binder material for different molar concentration of NaOH such as 8M, 10M and 12M. The compressive strength data obtained experimentally are assessed for GOF tests using EasyFit Software. Goodness-offit (GOF) tests such as Kolmogorov–Smirnov, KolmogorovSmirnov- Lilliefors, and Anderson–Darling tests on compressive strength data revealed that geopolymer mortars of all types of considered molarity follow the four-parameter “Johnson SB” probability distribution function. To understand uncertainty associated with geopolymer mortar different plots such as PDF, CDF, Survival function, Hazard Function and Cumulative Hazard function, P-P and Q-Q plots are drawn. Then a dataset of 250 observations from published literature is developed using LightGBM and Random Forest Model for sensitivity analysis. LightGBM demonstrated superior performance over Random Forest regressors in predicting the compressive strength of GP mortar, with an R2 of 0.94. Sensitivity analysis conducted using SHAP and partial dependence plots identified fly ash content as the most influential parameter affecting compressive strength, followed by the fine aggregate content, Na2SiO3/NaOH ratio, GGBFS content, NaOH molarity, and curing duration. The high sensitivity to fly ash is attributed to its dependency on thermal curing, where even small fluctuations in content can lead to significant strength variation. In contrast, GGBFS contributes to greater stability due to its ability to promote ambient-temperature curing. Designers and practitioners should prioritize these factors during the development of geopolymer mortars to minimize variability.