Prediction and deployment of compressive strength of high-performance concrete using ensemble learning techniques

Concrete is widely utilized in construction; however, accurately predicting its compressive strength is difficult due to the complex relationships within its mixture. Although previous studies have predicted the compressive strength of high-performance concrete (HPC), the literature lacks an optimized end-to-end framework for this purpose. To address this gap, this study conducts four experiments focusing on various stages of the predictive modeling process. Experiment 1 proposes and evaluates ensemble models using voting and stacking techniques and benchmarks them against eight other models. The experiment also investigates the impact of different data-splitting ratios on model performance. In experiment 2, these models are ranked based on ten evaluation metrics using the ELimination Et Choix Traduisant la REalité (ELECTRE) method. Experiment 3 involves selecting the best features for model retraining using Recursive Feature Elimination (RFE) to enhance predictive accuracy and efficiency. In experiment 4, the interpretability of the selected model is investigated using intrinsic and extrinsic methods. The results demonstrate that data splitting ratios influence model performance, as the ensemble stacking regressor using 30 % of the data for testing (ESR-30) outperforms other models. RFE improves the model's performance, reducing the root mean square error (RMSE) from 4.136 to 3.928 by 5.0 %, increasing the coefficient of determination (R²) from 0.937 to 0.943, and reducing the computational time by 30.8 %. Intrinsic feature importance and SHapley Additive exPlanations (SHAP) values identify age of testing and cement content as the most critical features influencing compressive strength predictions. The optimized model is deployed as a web-based application, providing a user-friendly interface for predicting HPC compressive strength. This study has significant implications for improving the reliability, efficiency, and accessibility of concrete compressive strength predictions in the construction industry.