Classification of multiple Dirichlet observations under a Multinomial Model

•Classification of compositional data for discriminating between classes of low to high response values•Application of the Dirichlet distribution for classifying compositional data into distinctive groups of performance•A methodology for the classification of coal properties and size distributions for a commercial coal processing facility

The amount of gas produced from a coal gasification facility depends crucially on the properties and the size distribution of the coal being used in the process. The particle size distribution and the composition of the coal are measured as compositional data. In this paper we apply the Dirichlet distribution for the inputs and present a new classification scheme for yielding low, medium and high gas production. The approach presented is a linear partitioning of the Dirichlet simplexes that can also be extended to high dimensional cases. An alternative clustering approach based on a distance measure is also presented.