AN INTERPRETABLE MODEL TO PREDICT THE SEISMIC FAILURE MODE OF RC SHEAR WALLS

The use of machine learning techniques in structural engineering has become widely popular for constructing predictive models that aim to understand the behavior of structures during seismic events. Such models often employ sophisticated methodologies to achieve high accuracy in decision-making. However, the interpretability of these predictive models is as crucial as their accuracy as engineers need insight into the decision-making process of the model to ensure its practicality. This research aims to achieve both transparency and accuracy by introducing an intelligible classification model designed to predict the potential seismic failure mode of reinforced concrete shear walls. To accomplish this, eight different machine learning methods are employed, using experimental failure modes of conventional shear walls as outputs and wall design parameters (such as compressive strength of concrete and axial load ratio) as inputs. The results indicate that the Decision Tree approach is the most suitable classifier, effectively providing both high classification accuracy and interpretability.