Explainable Machine Learning Model for Predicting Drift Capacity of Reinforced Concrete Walls

The ability to predict the drift capacity of reinforced concrete structural walls is critical to the seismic design process. The accuracy of such predictions has implications for construction costs, seismic safety, and reliability. However, the inability of an empirical model to capture any nonlinearity that exists between the drift capacity and different influencing variables can negatively impact the predictive performance. This study proposes a drift capacity prediction model for special structural walls based on the extreme gradient boosting machine-learning algorithm and a data set of 164 special boundary element wall tests. The efficiency of the proposed model is evaluated using a nested cross-validation approach, and the results reveal its superior predictive capabilities relative to the empirical equation adopted in ACI 318-19. To overcome the lack of interpretability of the model, SHapley Additive exPlanations are used to examine the relative individual and interactive effects of the different input variables on the drift capacity. Keywords: artificial intelligence; drift capacity; extreme gradient boosting; machine learning; reinforced concrete walls; special boundary elements.