Machine learning prediction of web-crippling strength in cold-formed steel beams with staggered slotted perforations

The application of staggered slotted perforations in cold-formed steel (CFS) members is increasingly prominent in modern construction. Understanding the web-crippling strength of CFS beams, especially those with staggered slotted perforations, is crucial in structural engineering. This study employs machine learning (ML) models to predict the web-crippling strength of these beams under one-flange loading conditions, specifically interior-one-flange and end-one-flange loading. The research utilises a comprehensive dataset comprising 576 web-crippling strength results obtained through numerical modelling. The dataset includes parameters such as yield strength, thickness, corner radius, slot length, slot width, and bearing plate length. Four different ML algorithms—k-nearest neighbour (KNN), random forest (RF), support vector regression (SVR), and artificial neural network (ANN)—are developed and evaluated. Performance metrics, including coefficient of determination (R²), mean squared error (MSE), root mean square error (RMSE), mean absolute error (MAE) and mean normalised bias (MNB) are used to assess model accuracy. The random forest model outperforms others in both the training and testing phases. Shapley additive explanation (SHAP) and partial dependence plots further analyse the influence of input features on web crippling strength. This study presents a robust ML-based approach for predicting web crippling strength, providing engineers with a time-efficient alternate method.