Predicting the uniaxial compressive and tensile strengths of ultra-high-performance concrete using super learner models

The aim of this research was to develop robust models for estimating the uniaxial compressive and tensile strengths of ultra-high-performance fibre-reinforced concrete. A total of 702 experimental compressive test results and 192 direct tensile test results were gathered to establish four base machine learning models (a decision tree, a random forest, an artificial neural network and extreme gradient boosting). Subsequently, super learner models (SLMs) were constructed to enhance predictive capability by integrating the outputs of the base models. An independent experimental programme was conducted to validate the predictive capability of the SLMs. The investigation revealed that the SLMs outperformed the base models and previously reported models in the literature, achieving R² values exceeding 0.982 for compressive strength and 0.904 for tensile strength, along with root mean squared errors below 5.062 MPa and 0.760 MPa, respectively. A graphical user interface incorporating the SLMs was developed, providing a practical and user-friendly tool for linking the compressive and tensile strength predictions. Sensitivity assessment indicated that concrete age and fibre volume content were the most influential factors affecting compressive strength and tensile strength, respectively. Finally, the accuracy and reliability of the SLMs were further confirmed through excellent agreements with independent experimental datasets.