Resumen
Assessing the seismic vulnerability of large numbers of buildings is an expensive and time-consuming task, requiring the collection of highly complex and multifaceted data on building characteristics and the use of sophisticated computational models. This study reports on the development of a data mining technique: Support Vector Machine (SVM) for resolving such multi-dimensional data problems for assessing buildings? seismic vulnerability at a regional scale. Particularly, we developed an SVM model for rapid assessment of the macroscale seismic vulnerability of buildings in terms of spectral yield and ultimate points of their capacity curves. Two case studies, one with 11 building characteristics and the other with 20, were used to test the proposed SVM model. The results show that when 20 building characteristics are included, an individual building?s seismic vulnerability in term of its spectral yield and ultimate points can be predicted by the proposed SVM model with an average 64% accuracy if the training dataset contains 400 samples, rising to 74% with 4400 training samples. Coupling the proposed technique with demand curves based on buildings? locations will enable rapid and reliable seismic-risk assessment at a regional scale, requiring only basic building characteristics rather than complex computational models.