Prediction of Skeleton Curves for Seismically Damaged RC Columns Based on a Data-Driven Machine-Learning Approach

The skeleton curve plays a crucial role in evaluating the seismic capacity of damaged structures. The research explored the application of data-driven machine learning approaches to predict the skeleton curves of earthquake-damaged reinforced concrete (RC) columns. Various machine learning methods, including Lasso regression, K-nearest neighbor (KNN), support vector machine (SVM), decision tree, and AdaBoost, were employed to develop a machine learning prediction model (MLPM) for seismic-damaged RC columns. A substantial dataset for the MLPM was derived from finite element (FE) analysis results. The input parameters for the machine learning models included the design specifications of the numerical column model and the damage index (DI), while the coordinates of key points on the skeleton curves served as the output parameters. The findings indicated that the K-nearest neighbor algorithm exhibited the best predictive performance, particularly for the yielding and peak points. The most influential input feature for predicting peak strength was the shear span-to-effective depth ratio, followed by the DI. The ML-based models demonstrated higher efficiency than numerical simulations and theoretical calculations in predicting the skeleton curves of damaged RC columns.