Investigation of Ensemble Machine Learning Models for Estimating the Ultimate Strain of FRP-Confined Concrete Columns

Accurately predicting the ultimate strain of fiber-reinforced polymer (FRP)-confined concrete columns is essential for the widespread application of FRP in strengthening reinforced concrete (RC) columns. This study comprehensively investigates the performance of ensemble machine learning (ML) models in estimating the ultimate strain of FRP-confined concrete (FRP-CC) columns. A dataset of 547 test results of the ultimate strain of FRP-CC columns was collected from the literature for training and testing ML models. The four best single ML models were used to develop ensemble models employing voting, stacking and bagging techniques. The performance of the ensemble models was compared with 10 single ML and 11 empirical strain models. The study results revealed that the single ML models yielded good agreement between the estimated ultimate strain and the test results, with the best single ML models being the K-Star, k-Nearest Neighbor (k-NN) and Decision Table (DT) models. The three best ensemble models, a stacking-based ensemble model comprising K-Star, k-NN and DT models; a stacking-based ensemble model comprising K-Star and k-NN models and a voting-based ensemble model comprising K-Star and k-NN models, achieved higher estimation accuracy than the best single ML model in estimating the strain capacity of FRP-CC columns.