Holistic model validation and interpretation with SHAP for machine learning models: predicting compressive strength of fly ash geopolymer concrete using optimization algorithms and ensemble model

This research lies in the application of a multi-model approach using explainable artificial intelligence (XAI) to interpret machine learning (ML) predictions for compressive strength (CS) of Fly Ash Geopolymer Concrete (FAGPC). The research utilized the use of 624 datasets collected from 10 published literature using four optimized models (Random Forest “RF”, Optimizable Decision Tree “ODT”, Optimizable support vector machine “OSVM”, and Optimizable Gaussian process regression “OGPR”). The curing age (A Day), Molarity (M), Fly ash (FA kg/m³), coarse aggregate (C kg/m³), temperature (T °C), fine aggregate (F kg/m³), superplasticizer (SP), NaOH/Na₂SiO₃ were among the input and CS (MPa) as the target variable. The study’s conclusions show that the OGPR-M2 model performed better than any of the other models with a high degree of precision. During the calibration phase, the mean absolute percentage error (MAPE) was discovered to be 9.428%, and the linear correlation coefficient (R) was computed to be 0.9705 respectively during the conventional ML techniques. Additionally, during the multi-model approach of Holistic-XAI, it was found that ODT surpassed all other models with a R and PCC equal 0.9929 and 0.9915 in calibration and verification phase respectively. The XAI analysis improved the transparency and dependability of ML predictions by offering interpretable insights into how each mix parameter, especially molarity, affects CS.