Advanced predictive modeling of shear strength in stainless-steel column web panels using explainable AI insights

• First study on shear strength of H-shaped stainless steel column web panels. • Existing design codes like EC3, focused on carbon steel is critically examined. • A machine learning-based methodology for stainless-steel panel zone is presented. • SHAP explainable AI to identify factors influencing panel zone behavior is used. • GUI for predicting shear strength of stainless steel column web panels is developed. In steel moment-resisting frames, energy dissipation occurs through yielding at the beam ends. Furthermore, the column panel zone can be designed to contribute to this energy dissipation process. The European standard (EN 1993–1–4) for stainless-steel is developed based on carbon steel procedures, without taking into account stainless steel's unique strain hardening and mechanical properties. This discrepancy may result in inaccuracies in predicting panel zone behavior. However, with the recent advancements in stainless steel, it is timely to reassess these limitations. The present research investigates the behavior of stainless-steel column web panels through an explainable artifactual intelligence methodology. This approach combines twelve widely recognized machine learning algorithms with the SHAP algorithm for enhanced explainability and transparency. In addition, a user-friendly graphical user interface has been developed to simplify engineering design. The Extra Trees Regression algorithm demonstrated the highest predictive performance, achieving R² = 0.987, mean absolute error (MAE) = 3.575 kN, and root mean square error (RMSE) = 6.464 kN for the entire dataset. The SHAP analysis revealed that bolt diameter and the column second moment of inertia are the most critical input features affecting shear strength. This approach effectively captures the nonlinear characteristics of shear behavior in stainless-steel column web panels and offers clear insights into the contribution of different factors. The developed method not only improves predictive accuracy but also promotes transparency, making it a practical tool for engineers in structural component design.