Zaker Esteghamati, MohsenFlint, Madeleine M.Rodriguez-Marek, Adrian2022-08-102022-08-102022http://hdl.handle.net/10919/111499Data incompleteness and uncertainty impede the application of performance-based design of structures at early design, which relies on data- and time-intensive numerical simulations. Early design is the most influential stage in a buildings' life cycle performance, hence neglecting quantitative methods to evaluate the design in preliminary stages can lead to missing on opportunities to improve building resiliency. This study presents a framework to implement surrogate models for supporting performance-based early design of concrete multi-story buildings. Five different surrogate models including multiple linear regression, random forest, extreme gradient boosting, support vector regression machines, and k-nearest neighbors are developed and compared to represent the seismic-induced structural loss of 720 generic concrete office buildings using early design parameters. Additionally, variance-based sensitivity is used to determine influential parameters for the best-performing model. The results show that extreme gradient boosting and support vector regression machines can be used to relate crude topology and design parameters to building seismic performance with reasonable accuracy.application/pdfenAttribution-NonCommercial-ShareAlike 4.0 InternationalSurrogate modelingEarly designPerformance-based engineeringMachine learningConcrete framesConference proceeding