Formulation of Shear Strength for Concrete Members Using SHAP and Iterated Local Search with Semantic Cluster Operator

With the rapid development of computer science, a lot of research has recently been done to propose shear strength prediction models for concrete members using machine learning algorithms. However, most of the existing studies were based on black box models. The problem with such an approach was that it was difficult to identify which process the models had undergone to make predictions. This, in turn, poses limitations in their use in the field of structural engineering that requires simple but mechanically explanatory models. In this regard, this study proposes shear strength equations with the use of an iterated local search using a semantic cluster operator (ILSS) capable of deriving a prediction model in the form of a simple equation. To this end, a total of 1,665 shear test results were collected for modeling, including those from reinforced concrete (RC), steel fiber-reinforced concrete (SFRC), prestressed concrete (PSC), and steel fiber-reinforced-prestressed concrete (SFRPSC) members. The proposed equations were validated for accuracy, accompanied by analyses of feature importance, feature effects, contribution rate of each term, and the strength reduction factor. The results revealed that the proposed models evaluated the shear strength of the collected specimens with a higher level of accuracy than the current codes and empirical equations proposed by existing studies. Furthermore, the shear strength results with the appropriate safety factor secured could be obtained by applying a strength reduction factor of 0.75.