| Damage response of Reinforced Concrete (R/C) buildings is a multiparametric issue that depends on structural characteristics and earthquake features. The distribution of masonry infills is one of the structural characteristics that can be optimized during the design phase of the buildings. Its use constitutes a widespread practice in many earthquake-prone regions around the world. This paper introduces Machine Learning (ML) techniques aiming to evaluate the impact of in-plan irregularities (caused by masonry infills) on the R/C buildings’ damage response. The training dataset has been developed by means of Nonlinear Time History Analyses of an R/C building with different masonry infills’ distributions, subjected to 65 real strong motions. The structural damage is expressed in terms of the Maximum Interstory Drift Ratio (MIDR). Six robust ML algorithms were utilized for this task, namely: Decision Trees, Random Forests, Artificial Neural Networks, Gradient Boosting Machines, k-Nearest Neighbors, and Support Vector Machines. Training was performed on the natural logarithm of MIDR, to ensure that the final predictions remain strictly positive after the application of the inverse exponential transformation. All of the developed models achieved very high performance, with a coefficient of determination higher than 0.93. Furthermore, the two best performing , had a mean error very close to 0, indicating their unbiased potential. Overall, the results demonstrate that the obtained robust models should be utilized to obtain reliable predictions for this complex and impactful task. |
*** 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.