Strength Prediction of Ultra-High Performance Concrete (UHPC) Based on BOHB-XGBOOST Algorithm

Ling Wang^1,2, Mohammad Faizuddin Md Noor^2,*, Yanan Zhang^3,*
1 School of Artificial Intelligence and Electronic Information, Nantong Vocational University, Nantong, China
2 Malaysian Institute of Information Technology, Universiti Kuala Lumpur, Kuala Lumpur, Malaysia
3 College of Water Conservancy and Hydropower Engineering, Hohai University, NanJing, China
* Corresponding Author: Mohammad Faizuddin Md Noor. Email: email ; Yanan Zhang. Email: email
(This article belongs to the Special Issue: Emerging Artificial Intelligence & Data-Driven Modeling in Civil Engineering)

Computer Modeling in Engineering & Sciences https://doi.org/10.32604/cmes.2026.084489

Received 23 April 2026; Accepted 27 July 2026; Published online 10 August 2026

Download PDF

Abstract

Ultra-high-performance concrete (UHPC) relies on multivariable mix design and curing regimes, which makes empirical estimation of compressive strength increasingly unreliable when material systems vary. In this context, unlike previous studies that only applied standard eXtreme Gradient Boosting (XGBoost), this study introduces an advanced hybrid optimization strategy, Bayesian Optimization and Hyperband (BOHB), which combines the sample efficiency of Bayesian optimization with the resource allocation mechanism of Hyperband, and incorporates SHapley Additive exPlanations (SHAP) for influencing factor analysis, thereby proposing a BOHB-XGBoost framework integrated with SHAP analysis. The proposed model demonstrates excellent predictive accuracy and stability, achieving a coefficient of determination R² of 0.967 and a Mean Absolute Percentage Error (MAPE) of 4.86% on the test set. Comparative experimental results indicate that this model shows improved and more stable performance than mainstream machine learning methods, including standard XGBOOST, Random Forest, Gradient Boosting Decision Trees, Support Vector Regression and a Multilayer Perceptron across multiple evaluation metrics. This fully demonstrates the significant advantages of the BOHB optimization framework in processing structured tabular data. By balancing exploration and exploitation during the hyperparameter search process, this strategy effectively overcomes the limitations of traditional manual tuning and significantly enhances the model’s generalization capability. Furthermore, the SHAP-based interpretability analysis identifies the key factors influencing UHPC strength, such as the water-to-cement ratio and steel fiber content. These findings are highly consistent with the experimental results conducted in this study, thereby validating the model’s physical plausibility and engineering applicability. This research not only provides an efficient and reliable method for predicting UHPC strength but also offers a transparent and interpretable scientific basis for the rational optimization of mix proportion design.