Machine learning approach for predicting the compressive strength of ultra-high performance fiber reinforced concrete (UHPFRC)

Ultra-High Performance Fiber Reinforced Concrete (UHPFRC) is an advanced cementitious composite which contains fibers. UHPFRC possesses improved mechanical properties, durability, workability, fire resistance, abrasion resistance, and low permeability. As a result, it finds extensive applications in casting full-scale structural elements and for structural retrofitting purposes. This research focuses on predicting the compressive strength of UHPFRC by employing Machine Learning (ML) techniques. Initially, a comprehensive literature review was undertaken to extract mix design details from previous experimental studies and a database was developed with 200 data points. Seven ML models, including Support Vector Regressor, Decision Tree Regressor, Random Forest Regressor, Gradient Boosting Regressor, Light Gradient Boosting Regressor, Extreme Gradient Boosting Regressor, and Multi-Layer Perceptron Neural Network were constructed for the estimation of compressive strength. Ten input parameters representing different material properties were utilized by the models, while the output parameter was the compressive strength of UHPFRC. The models underwent performance evaluation through the computation of various performance evaluation parameters. Among these models, the XGBR model demonstrated the highest prediction accuracy with an R² value of 0.905 and MSE value of 69.48 and hence was selected for detailed analysis. Overall, boosting-based models outperformed the rest and the SVR model showed the least accuracy. Further, a SHAP (Shapley Additive Explanations) analysis was conducted to unveil the black box nature of the ML model and to provide more detailed interpretations. A feature importance analysis was undertaken based on mean absolute SHAP values to investigate the impact of each parameter on the performance of the material. In addition, guidelines for the utilization of material parameters were presented at the conclusion to achieve optimal compressive strength of UHPFRC and provide a practical framework for UHPFRC mix design.