Predicting the Compressive Strength of UHPFRC Using Machine Learning and Soft Computing Models: Optimization of Fiber Content and Additives

Ultra-high-performance fiber-reinforced concrete (UHPFRC) is a relatively new material known for its superior mechanical properties, particularly its compressive strength (CS), making it suitable for advanced structural applications. Traditional experimental methods for predicting CS are time-consuming and costly. In this study, a dataset of 276 samples with 12 input parameters was compiled from existing literature to develop predictive analytical models. The input variables include cement, sand, water, superplasticizer, silica fume, fiber content, water–binder ratio, water–cement ratio, curing age, fiber aspect ratio, temperature, and fiber volume. The reported CS values range from 90 to 186 MPa. Five modeling techniques—Linear Regression (LR), Log Base Regression (LBR), Nonlinear Regression (NLR), M5P-tree, and Artificial Neural Network (ANN)—were employed to predict the compressive strength of UHPFRC. Among these models, ANN demonstrated the highest prediction accuracy across all evaluation criteria, followed by the M5P-tree model. Residual error analysis confirmed that the ANN produced the lowest prediction error. Sensitivity analysis revealed that temperature, curing age, and superplasticizer content significantly influence CS. Optimization results indicated that a fiber content between 2.05% and 2.09% yields maximum compressive strength. These findings provide valuable insights for optimizing UHPFRC mix design using machine learning approaches.