LEVERAGING ADVANCED ENSEMBLE LEARNING FOR STATE-OF-THE-ART PREDICTION OF CONCRETE COMPRESSIVE STRENGTH

Abstract

The accurate prediction of concrete compressive strength is crucial for structural design and quality control. This study investigates the application of machine learning models to forecast strength based on mixture composition and age, comparing the performance of Extreme Gradient Boosting (XGBoost) and Support Vector Machine (SVM) algorithms. Methods: A dataset of 1,030 instances with eight input features (cement, slag, water, etc.) and one target (Strength) was utilized. Following data preprocessing and correlation analysis, both models were developed and rigorously tuned. Their performance was evaluated using k-fold cross-validation and metrics including R², RMSE, and MAE. Results: XGBoost demonstrated superior predictive capability, achieving an R² of 0.939, an RMSE of 4.11 MPa, and an MAE of 2.74 MPa. In contrast, SVM performed less effectively, with an R² of 0.880 and higher errors (RMSE: 5.79 MPa, MAE: 4.07 MPa). Violin plots confirmed XGBoost lower variance and absence of erroneous predictions compared to SVM. Feature importance analysis identified cement content, superplasticizer, and age as the most statistically significant factors. Discussion: The results conclusively establish XGBoost as the more robust and accurate model for this task, effectively capturing the complex, non-linear relationships in the data. It is recommended for deployment in predictive tasks to optimize concrete mix designs and reduce reliance on destructive testing.