AI based predictive modelling for compressive strength of metakaolin-based geopolymer concrete incorporated with Nano Titanium

Metakaolin-based geopolymer concrete (GPC) has gained significant attention as a sustainable alternative to conventional concrete due to its superior mechanical properties, reduced carbon footprint, and potential for incorporating industrial by-products. This study evaluates the performance of nine artificial intelligence (AI) techniques in predicting the compressive strength (CS) of metakaolin-based GPC. The input parameters considered were the contents of metakaolin, nano titanium (NT), metakaolin to fine aggregate ratio, metakaolin to coarse aggregate ratio, metakaolin to total aggregate ratio, sodium silicate, sodium hydroxide to sodium silicate ratio, superplasticizer, and curing days. At the same time, the CS served as the output parameter. The comparative analysis revealed significant variations in the prediction accuracy of the models, highlighting their potential applicability and limitations in modeling the CS of GPC. This research provides valuable insights for selecting optimal AI techniques in the domain of CS prediction. XG boost regression model performed well in predicting the CS of GPC with R2 0.9954, with an RMSE of 0.64 MPa, which outperformed all other models tested. The model interpretation through SHAP analysis showed curing days as the strongest predictor of strength while longer duration positively affected the outcomes.