Bayesian Approach to the Classification of BMI Time Series Data from Babyhood to Junior High School Age of Japanese Children

The time developments of BMIs (Body Math Index) of children are known to be classified into several types, and the results can be utilized to their health guidance. For this purpose, we approach to the classification problem of the time developments of the BMI data of Japanese children from their babyhood to junior high school age. We have inferred the dimension of their principal component space and the number of component distributions of a Gaussian mixture model, adopting a framework of variational Bayesian statistical inference. As a result, the data are found to be classified into 8 types in a 12-dimensional principal component space.