A1221
Title: Feature selection in censored-covariate regression models
Authors: Garth Tarr - University of Sydney (Australia) [presenting]
Rajan Shankar - University of Sydney (Australia)
Tanya Garcia - University of North Carolina at Chapel Hill (United States)
John Ormerod - The University of Sydney (Australia)
Abstract: Feature selection in regression models involving a continuous response variable, y, and a right-censored time-to-event covariate, x, is studied. Since different features may be associated with y and x, respectively, a regularised approach is developed that allows distinct feature selection in the two model components. Estimation is complicated by censoring in x, so an expectation-maximisation (EM) algorithm is used that alternates between imputing censored values and updating model parameters. Tuning parameter selection and computational implementation are discussed, and the method is illustrated using Huntingtons disease data. More generally, the proposed framework provides a practical approach to feature selection in regression problems with censored covariates.
