ML Model Hybridization: Concrete Strength Prediction

In this study, we use Hybridized Machine Learning (ML) algorithms for prediction of Compressive Strength (CS) of fly-ash based Geopolymer Concrete (GPC). The base models include Extreme Gradient Boosting (XGB) and Random Forest (RF). The hybridization involves using metaheuristic optimization algorithms (like Particle Swarm Optimization - PSO and Grey Wolf Optimization - GWO) to tune the models’ hyperparameters. A 5×5 nested cross-validation (nested CV) approach was adopted to ensure unbiased hyperparameter tuning and reliable evaluation of model performance.

Model performance was evaluated using statistical indices and sensitivity analysis. The findings showed that the PSO-XGBoost model outperformed other models with accuracy of R2 and RMSE of 0.97 and 3.85 MPa respectively. The ranking of the ML models obtained was: GWO-XGBoost > PSO-Random Forest > GWO-SVR for both train and test set. SHAP analysis identified Curing Temperature and Curing Time as the most critical parameters effecting CS.

Overall Procedure for Hybridized ML Model Development

Radar plot showing model performance of a) Training Set b) Testing Set

SHAP Importance Plot for GPC Compressive Strength

PSO-XGB (Best perrforming Model)

Paper On Review!
Soon