BAYESIAN WAVELET-BASED CURVE CLASSIFICATION VIA DISCRIMINANT ANALYSIS WITH MARKOV RANDOM TREE PRIORS.

摘要: Discriminant analysis is an effective tool for the classification of experi- mental units into groups. When number variables much larger than observations it necessary to include a dimension reduction procedure in inferential process. Here we present typical example from chemometrics that deals with different types food species via near infrared spectroscopy. We take nonparametric approach by modeling func- tional predictors wavelet transforms and then apply discriminant domain. consider Bayesian conjugate normal model, ei- ther linear or quadratic, avoids independence assumptions among coefficients. introduce latent binary indicators selection discrimi- natory coefficients propose prior formulations use Markov random tree (MRT) priors map scale-location connections wavelets conduct posterior inference MCMC methods, show performances on our case study authenticity, compare results several other procedures.