Abstract
We consider the problem of discriminating some multiple stationary time series observations into one of two mutually exclusive and exhausitive populations. A discriminant procedure is constructed on the Fourier transformed data as that of“zero mean difference”in ordinary multivariate analysis. Performances of the procedure are discussed based on the probabilities of misdiscrimination for which three approximate formula are investigated, and on a measure of performance which is also derived from J-divergence. Finally, we illustrate the procedure using two artificial time series data on autoregressive model and give some suggestions for its applications.
