Study on the Compressive Strength Predicting of Steel Fiber Reinforced Concrete Based on an Interpretable Deep Learning Method

Steel fiber reinforced concrete (SFRC) exhibits excellent material enhancement and toughening properties. It is widely used in applications such as airport runways, highway pavements, and bridge deck overlays. In order to predict the compressive strength of SFRC efficiently and accurately, this study proposes a deep learning-based prediction model, trained and tested on a large set of experimental data. Additionally, the SHapley Additive exPlanations (SHAP) interpretability method is employed to analyze and interpret the prediction outcomes. SHAP facilitates the identification and visualization of both positive and negative correlations among input features, along with their magnitudes and overall importance from local and global perspectives. This analysis sheds light on the decision-making logic of the “black-box” model and addresses the transparency challenges typically associated with conventional machine learning (ML) approaches. Fourteen physical parameters, including steel fiber content, length, diameter, cement dosage, coarse aggregate content, and fly ash content, are selected as input features. The SHAP values of these parameters are visualized to assess their importance, impact, and influencing patterns on compressive strength prediction. The results show that the optimized deep learning model has higher prediction accuracy and generalization ability compared to other traditional ML models. The SHAP analysis results are consistent with the experimental results, and the predictive model well reflects the complex nonlinear relationship between various characteristic parameters, which can provide a basis and reference for the engineering design of SFRC materials.