Machine Learning (ML) driven classification and optimization of silica fume- polypropylene fiber induced geopolymer concrete using clustering and regression techniques

Geopolymer concrete, made from fly ash and Ground Granulated Blast Furnace Slag (GGBS), is being considered a promising sustainable replacement for ordinary cement-based concrete, especially under ambient curing conditions. But its brittle nature restricts its usage. The present study aims to investigate the improvement in the mechanical properties, such as compressive strength, split tensile strength, and flexural strength of geopolymer concrete by the addition of polypropylene fibers (PPF) and silica fume (SF). The binders used for the study are a blend of 50% fly ash and 50% GGBS to which silica fume was added at 5%, 10%, and 15%by weight of binder, and polypropylene fibers in varying fractions by volume of mix, ranging between 0% and 1% with an increment of 0.25%. A fixed alkaline activator of 12 M NaOH and sodium silicate was used. After 7 & 28 days of ambient curing, the mechanical strength tests were carried out on the fiber-reinforced geopolymer concrete specimens. The findings showed that the addition of polypropylene fibers and silica fume significantly improved the strength characteristics. Optimal performance is obtained at a 0.5% Polypropylene fiber content and a 15% replacement of binder content (FA+GGBS) with silica fume. These experimental results were used to prepare the data sets for further analysis using clustering techniques and predictive models. Based on the clustering analysis, the mixes were categorized as high-performance, economical, and early-strength mixes. Later, supervised machine learning models such as Random Forest and XGBoost are employed to predict the strength of fiber-reinforced geopolymer concrete. Based on the statistical performance metrics, the Random Forest model performed better has shown a reliable performance with R²(test) = 0.928 and lower RMSE = 4.258 for compressive strength when compared to XGBoost (R² (test)= 0.886, RMSE = 5.363). For splitting tensile strength (STS), Random Forest outperformed XGBoost with R²(test) = 0.810 and RMSE = 0.484, whereas XGBoost produced R²(test) = 0.741 and RMSE = 0.565. Flexural strength (FS), Random Forest again showed superior performance with R²(test) = 0.958 and RMSE = 0.339 compared to XGBoost (R²(test) = 0.904, RMSE = 0.511).