B0702
Title: Multinomial functional regression with application to lameness detection for horses
Authors: Helle Sorensen - University of Copenhagen (Denmark) [presenting]
Abstract: The motivation arises from a dataset consisting of 85 acceleration signals collected from trotting horses that are either healthy or have an induced lameness on one of the four limbs. Our aim is to develop a method that uses such a signal for detection of lameness and identification of the lame limb. This is a supervised classification problem with five groups and functions (acceleration curves) as predictors. We propose to use a multinomial functional regression model. We combine the discrete wavelet transform and LASSO penalization for estimation of the model and use the fitted model to predict the class membership for new curves.
