Data-Driven Bearing Capacity Prediction of Self-drilling Screw in Cold-Formed Steel Using Machine Learning

Many branches of civil engineering are utilizing the benefit of machine learning (ML) techniques for solving complex problems such as predicting the capacity of fiber-reinforced concrete or steel connections. This paper presents an application of machine learning techniques in bearing strength prediction of self-drilling screw connections in cold-formed steel. Cold-formed steel and its structural forms are getting more popularity and interest from stakeholders because of their easier application, simple fabrication, and lower cost. An experimental database comprising 278 specimens has been developed from the conducted tests and existing literature. The database contains different features of the connections, explicitly end and edge distances, screw diameter and numbers, plate width and thickness, ultimate yield and ultimate strengths of the plates, and pitches along and perpendicular to the loading directions. Three ML-based regression models, namely linear regression (LR), ridge regression (RR), and support vector machine (SVR) are selected for the bearing resistance prediction of screw connections. Those models’ performances are evaluated based on the coefficient of determination (R2), adjusted R2 (Adj. R2), root mean square error (RMSE), and mean absolute error (MAE). The result indicates that SVR performs the best among the proposed models, and the linear and ridge technique performs poorly. Connections bearing strengths predicted by the proposed models are also compared with the existing code-based formulas.