Hybrid Metaheuristic Optimized Random Forest Models for Predicting Compressive Strength of Alkali Activated Concrete

This study explores the application of hybrid machine learning models for predicting the compressive strength (CS) of alkali-activated concrete (AAC), a sustainable substitute for traditional Portland cement concrete. A random forest (RF) model was optimized using six metaheuristic algorithms: differential evolution (DEA), human felicity algorithm (HFA), nuclear reaction optimization (NRO), lightning search algorithm (LSA), Harris hawks optimization (HHO), and tunicate swarm algorithm (TSA). Among these, the NRO-RF model achieved the highest performance with an R² of 0.931, surpassing TSA-RF (0.918), HHO-RF (0.846), LSA-RF (0.828), HFA-RF (0.811), and DEA-RF (0.794). These hybrid models not only offered high predictive accuracy but also delivered stable and generalizable predictions across varied mix proportions, supporting more reliable AAC design and quality control. Interpretability techniques revealed that higher SiO₂/Na₂O ratios (S/N), sodium hydroxide (NaOH), and blast furnace slag ratio (BFSR) positively influenced CS, while excessive water (W), aggregate (Agg), and precursor content (PC) had negative effects. These insights provide practical guidance for optimizing mix designs. Additionally, a user-friendly graphical interface was developed to facilitate easy CS prediction, reducing reliance on physical testing and promoting efficient, data-driven decision-making in AAC development.