| Beam-column joints play one of the most crucial roles in the overall seismic behavior of reinforced concrete structures. Joints support the adjacent beams and slabs and thus, their failure could lead to loss of overall structural stability, partial or total collapse of the building, and loss of human lives. Hence, reliable estimation of these components is of great importance. This research effort, discusses the development of the following robust Machine Learning models, namely: Extra Trees, XGBoost, k-Nearest Neighbors, Support Vector Regression, and Gradient Boosting. Moreover, the Voting and Stacking ensemble methodologies were employed to offer optimization and generalization capabilities. The algorithms were trained on an extensive dataset comprised of 550 experimental measurements of both interior and exterior joints. Furthermore, 5-fold cross-validation was employed, aiming to reduce the variance of the performance metrics and to increase their reliability. All of the individual models achieved very high performance, with a coefficient of determination R2 as high as 0.84, and a Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) as low as 0.989 and 1.396, respectively. The two best performing models had a mean and median error close to 0. They also had approximately equal percentage of positive and negative errors, indicating their unbiasedness. The ensemble models maintained these attractive properties, while obtaining a slightly increased performance, with R2 = 0.85, MAE=0.938, and RMSE=1.355. Overall, the results indicate that the trained models can be confidently employed on this important engineering challenge. |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.