A soft-computing-based modeling approach for predicting acid resistance of waste-derived cementitious composites

This research aimed to build estimation models for the compressive strength (C-S) of cement mortar containing
eggshell and glass powder after the acid attack using machine learning algorithms. A lab test data comprising 234
data points with 8 input factors was utilised for modelling. Four ensemble machine learning techniques,
including gradient boosting, AdaBoost, random forest, and bagging, were employed to achieve the research’s
goals. In addition, to examine the influence and correlation of input factors, a SHapley Additive exExplanations
(SHAP) analysis was conducted. The built estimation models well agreed with the lab test results based on R2 and
the variance between actual and model estimated results (errors). Random forest and bagging exhibited superior
prediction performance, with R2 of 0.982 and 0.983, respectively, than gradient boosting and AdaBoost, with R2
of 0.969 and 0.977, respectively. The comparative analysis of statistical measures also indicated superior accuracy
of random forest and bagging, with mean absolute percentage error (MAPE) of 2.40%, than gradient
boosting and AdaBoost, with MAPE of 2.90% and 2.60%, respectively. SHAP analysis exhibited that the highly
influential factor for the acid resistance of glass and eggshell-based mortar was the 90-day C-S of the sample,
followed by the quantity of glass powder, eggshell powder, sand, cement, water, superplasticizer, and silica
fume.