TOWARDS A LONG-MEMORY-BASED MODEL FOR PREDICTION OF HEMOGRAM EVOLUTION

摘要

Time series of hematocrit values recorded daily over a period of 1024 days are analyzed as samples of a stochastic process. Methods for estimating the parameters of the stochastic model are compared. A linear predictor based on a fractional ARIMA process is introduced. A comparison between the empirical hematocrit series and their corresponding predictions, using selected values of the stochastic model parameters (d, κ and τ), indicates a close adequacy between them. Further characterization of others hemogram components could significantly contribute to design a theoretical model to predict whether hemogram series evolve towards normality or in the opposite sense.