Developing a robust machine learning framework for predicting the behavior of large-scale structure

A rapid and accurate analysis is crucial to structural design and control. This paper presents a comprehensive and robust framework for the analysis and prediction of the behavior of largescale structures. To ensure accurate prediction of structural responses, the proposed framework integrates sample selection, dimensionality reduction, and advanced machine learning analysis techniques. Three different types of structures are used as numerical examples: a tower truss structure, a steel building, and a reinforced concrete structure. The framework is trained using 26 different machine-learning methods and validated using a comprehensive set of performance metrics. Additionally, an enhanced machine learning method is introduced to achieve more accurate results. This technique leverages chaos game optimization to automate the parameter updating of the machine learning method. Shapley Additive Explanations (SHAP), as the interpretability technique, was incorporated into the framework to quantify each feature's contribution, helping engineers identify key factors influencing structural behavior and ensuring safer, more efficient designs. The validation results demonstrate the high accuracy of the proposed framework in predicting the behavior of large-scale structures. The implications for structural engineering are significant, as this framework has the potential to enhance the design and assessment of large-scale structures, thereby enhancing their safety, durability, and performance.