Optimizing Gene Expression Programming to Predict Shear Capacity in Corrugated Web Steel Beams

Corrugated web steel systems, such as corrugated web girders (CWG) and beams (CWSB), have the potential to influence the modern construction industry due to their unique properties, including enhanced shear strength and reduced necessity for transverse stiffeners. Nevertheless, the lack of a rapid and accurate design approach still limits its wide applications. Recently, gene expression programming (GEP) has been employed to predict the shear capacity of cold-formed steel channels, demonstrating superior predictive accuracy and compliance with established standards. This study applies GEP to predict the shear capacity of sinusoidal CWSBs and optimizes its predictive performance by employing a systematic grid search to explore combinations of chromosomes, head sizes, gene counts, and linking functions. The process involved testing 19 different parameter combinations and more than 60 developed models. The findings include the sensitivity of the model's performance to gene count and the critical role of the linking function. The optimal model in the study, GEP13, achieved R² of 0.95, an RMSE of 100.5, and an MAE of 86.6 in the testing dataset with 150 chromosomes, a head size of 12, and four genes using a multiplication linking function.