Intelligent Data Driven Ensemble Approaches for Bending Strength Prediction

This article presents a research study published in the Bulletin of Computational Intelligence that focuses on developing a machine learning framework for predicting the bending strength of ultra-high performance fiber reinforced concrete (UHPFRC) beams. The study addresses the limitations of existing code equations, which often yield inaccurate predictions with high coefficients of variation. By compiling a database of 264 experimental UHPFRC beam tests, the authors partitioned the data into training, validation, and testing subsets. They optimized six ensemble algorithms, including Random Forest and Gradient Boosting, using Bayesian hyperparameter tuning. The results indicate that the best-performing models, CatBoost and XGBoost, achieved high predictive accuracy on unseen test data, significantly outperforming traditional code equations. The research also emphasizes the interpretability of the machine learning models, identifying key predictors of bending strength. This work aims to provide a reliable data-driven approach to complement existing design codes and enhance structural engineering practices.