Reliable determination of peak shear strength of H-shaped concrete squat walls using explainable machine learning techniques

Flanged reinforced concrete walls also known as H-shaped walls are frequently used in nuclear facilities and conventional buildings due to their substantial lateral strength and stiffness in both directions. These walls mostly fail in shear, and it is essential to accurately estimate their peak shear strength ($V_p$). However, the provisions of existing building codes to determine peak shear strength have significant limitations such as exclusion of the influence of flanges and consideration of insufficient input parameters. Therefore, this study aimed to construct predictive models for H-shaped walls using machine learning techniques like Bagging Regressor (BR), Gene Expression Programming (GEP), and Extreme Gradient Boosting (XGB), based on data gathered from existing literature. The gathered data had twelve inputs including shear span ratio ($h_w$/$l_w$), the ratio of flange thickness to web thickness ($t_f$/$t_w$), and loading type (M) etc. Out of all the algorithms, only GEP depicted its output as an equation. The performance of the algorithms was checked using error metrices like Objective Function (OF), and coefficient of determination ($R^2$) etc. which showed that XGB exhibited the highest accuracy having the testing $R^2$ equal to 0.99. Additionally, shapely (SHAP), Individual Conditional Expectation (ICE), and partial dependence plots (PDP) analysis were employed which showed that flange length, loading type, and shear span ratio are some of the most contributing variables to determine $V_p$. Furthermore, a graphical user interface (GUI) has been developed to efficiently compute $V_p$ of RC squat H-shaped walls to help professionals in the civil engineering industry.