Publication
Indonesia is located in a high-seismic-risk region with a significant number of non-engineered houses, which typically have a higher risk during earthquakes. Due to the wide variety of differences even among parameters within one building typology, it is difficult to capture the total risk of the population, as the typical structural engineering approach to understanding fragility involves tedious numerical modeling of individual buildings-which is computationally costly for a large population of buildings. This study uses a statistical learning technique based on Gaussian Process Regression (GPR) to build the family of fragility curves. The current research takes the column height and side length as the input variables, in which a linear analysis is used to calculate the failure probability. The GPR is then utilized to predict the fragility curve and the probability of collapse, given the data evaluated at the finite set of experimental design. The result shows that GPR can predict the fragility curve and the probability of collapse well, efficiently allowing rapid estimation of the population fragility curve and an individual prediction for a single building configuration. Most importantly, GPR also provides the uncertainty band associated with the prediction of the fragility curve, which is crucial information for real-world analysis.