Transparent AI for Structural Engineering: Explainable Shear Strength Prediction in SFRC Beams Without Transverse Reinforcement

Predicting the shear strength of steel fiber-reinforced concrete beams is crucial for the design and assessment of concrete structures incorporating steel fibers. Traditional empirical methods, developed from limited datasets, often attempt to provide dependable predictions of SFRC shear strength. Moreover, traditional methods fail to catch the complex non-linear behaviour of steel fiber reinforced concrete beams. Machine learning models can analyze large datasets and uncover complex relationships, offering a promising alternative. The main challenge with high-accuracy ML models lies in their lack of interpretability, as they often behave as black-box systems, making it difficult for engineers to understand how the model reaches its predictions. To overcome this, Shapley Additive exPlanations (SHAP) and partial dependence plot (PDP), a technique designed to provide transparency in machine learning models, were used to build an explainable ML model for predicting the shear strength of SFRC beams. The experimental & existing shear design models were compared to the developed ML model to evaluate the effectiveness of the model. The results indicated that the XGBoost (R² = 0.988) model made predictions very close to the experimentally observed shear strengths, offering better and more unbiased performance compared to existing models. The SHAP analysis revealed that the most influential factors affecting shear strength predictions were beam width and effective depth, followed by concrete strength and longitudinal reinforcement ratio. Steel fiber factor and the shear-span to effective depth ratio were also significant contributors. By employing XGBoost and integrating SHAP to explain the model’s decisions, this study provides an accurate and interpretable method for predicting shear strength in SFRC beams.