Data-informed design equation based on numerical modelling and interpretable machine learning for the shear capacity of cold-formed steel hollow flange beams with unstiffened and edge stiffened openings

This study introduced a novel approach to derive a design equation using Machine Learning (ML) to estimate the shear strength of Cold-Formed Steel (CFS) Rectangular Hollow Flange Beams (RHFBs) with unstiffened and edge-stiffened circular openings in web. Based on experimental data, a comprehensive series of numerical modelling was conducted to develop data for ML models. Machine learning models, namely, (a) Extreme Gradient Boost (b) Light Gradient Boosting Machine (c) Random Forest and (d) Support Vector Regressor were trained and validated using the data obtained from parametric numerical models. The XGB model showed the best predictive accuracy for shear strength. Additionally, Shapley Additive Explanations (SHAP) were used to interpret the model predictions. The developed ML models showcased better accuracy compared to the existing semi-empirical equations. As a novel complementary approach, this study developed a simplified equation for the first time using XGB and Shapley values to estimate the shear strength reduction factor of doubly symmetric RHFBs with both edge-stiffened and unstiffened circular openings. The proposed equation obtained a reliability index (β) of 2.78 and outperformed the existing semi-empirical formulations, resulting in an improved design formula for RHFBs.