Independent component regression for predicting the responses of biaxial base‐isolated buildings

This paper presents a regression model to predict the base displacement responses of biaxial base‐isolated buildings using independent component analysis. The model proposed utilizes multiple ground motion intensity measures from North American and Japanese earthquakes as inputs, and transforms them into an independent component space using independent component regression (ICR). Unlike other latent variable methods, such as principal component regression, ICR does not readily allow for dimensionality reduction of the components that do not contribute significantly to the explained variance of the original data set. Hence, a whitening‐step to transform the correlated variables into uncorrelated ones is introduced prior to performing ICR. Prediction results are presented and compared with the simulation results for two building models with increasing degree of complexity. The results show that the model based on ICR results in good estimates for the base displacement responses, and the standard errors remain relatively small and constant across a range of isolation periods. Copyright © 2009 John Wiley & Sons, Ltd.