Abstract Summary
This paper deals with data-driven-based regression models for the estimation of maximum inter-storey drift demands (MIDR) in multi-story cross-laminated timber (CLT) buildings. Buildings with a different number of storeys (6, 8, 12, 16, and 20) are defined to cover medium- to high-rise CLT buildings. Different levels of panel fragmentation are considered by dividing a wall into sub-panels. The buildings were also designed based on several behaviour factors (q) to provide different ductility levels. The predictors used in the models are informed by data-driven techniques based on a vast set of nonlinear response history analyses (NRHAs) of 69 CLT buildings subjected to strong earthquakes. We explore various prediction-based methods to develop the models, including Linear regression, Polynomial regressions, and Machine learning algorithms like Decision trees, K-nearest Neighbor, and Support Vector Regressors. The result indicates that Machine learning algorithms have superior prediction power compared with traditional regression models. The ML-based predictions agree more with the numerical results of NRHAs as reflected by the highest R2 and lowest RMSE for the MIDR estimates. Finally, the result obtained from this approach is compared and discussed with those coming from previous studies.
