Use of interpretable machine learning methods to predict the fundamental period of masonry infilled reinforced concrete frame structures

Abstract

Machine learning has been used in predicting the natural period of vibration of reinforced concrete structures. However, their lack of interpretability diminishes the end user’s trust in machine learning predictions. Alternatively, the authors employed four classical machine learning methods coupled with eXplainable Artificial Intelligence (XAI) to forecast the fundamental period of vibration of Reinforced Concrete structures with masonry infills. We used SHapley Additive explanations (SHAP) to interpret the models and their predictions. Our analysis indicated that Random Forest (R2=0.999) was the best predictive model. All four machine learning models were better than the existing equations provided in design standards. As the novelty, SHAP explanations revealed the reasoning behind machine learning predictions. Accordingly, the Number of Storeys and Opening Ratio are the most influential parameters that govern the natural period of reinforced concrete structures with masonry infill walls. Therefore, this study suggests that explainable machine learning can be effective in structural engineering applications to help decision-making.