A non-homogenous hidden Markov model for statistical evaluation of food functionality

In recent years, the mechanism on how food have influence on health have been of great interest from the medical standpoint. For this research, the measurements on medical and genic factors have been carried out through clinical trials and the statistical methods for evaluating food functionality have been considered. Standard methods such as linear regressions and statistical tests, however, may not necessarily contribute to reliable medical assessments, because of the difficulty in expressing the distributions of the medical data by a normal distribution. Our goal throughout this article is to develop a new method for expressing the distribution of medical data and its application to the evaluation of food functionality. More specifically, we deal with a prediction problem on time-varying distribution of H1 influenza virus antibody titer after ingesting mushroom. For prediction, we develop a model for expressing a mixture probability distribution on the antibody titer in the class of non-homogenous hidden Markov model. Our prediction experiments have shown that the presented method improves the prediction accuracy in the case of using a standard multiple regression model, which gave a justification of introducing the model structure.