Enhancing Compressive Strength Estimations of Rice Husk Ash Concrete Utilizing Metaheuristic Optimization Algorithms

This study employed hybrid machine learning models, known for their superior accuracy over traditional ML models, to develop a reliable framework for estimating the compressive strength (CS) of rice husk ash (RHA) concrete. This approach eliminates the reliance on time-consuming and resource-intensive physical experiments. To optimize the hyperparameters of the random forest (RF) model, six metaheuristic algorithms were employed. These include the differential evolution algorithm (DEA), human felicity algorithm (HFA), lightning search algorithm (LSA), nuclear reaction optimization (NRO), Harris hawk optimization (HHO), and tunicate swarm algorithm (TSA). To assess the efficacy of the suggested ML models, several statistical indicators were employed. To enhance the interpretability of the model predictions, the SHapley Additive exPlaination (SHAP) method and partial dependence plots (PDP) analysis were utilized. All six hybrid models showed strong performance, with LSA-RF proving to be the most effective. LSA-RF had the highest coefficient of determination (R2) value of 0.979, showcasing superior prediction accuracy in comparison to DEA-RF (0.972), HFA-RF (0.962), NRO-RF (0.960), TSA-RF (0.916), and HHO-RF (0.928). Furthermore, an intuitive user interface was designed for practical applications, enabling instant CS predictions for RHA concrete based on input parameters.