Empirical modeling of compositions in chemical engineering

Abstract

Many of the data used in chemical engineering are compositional. When mass balance based theoretical models cannot be used, and empirical statistical modeling is applied, the compositional, dependent, multivariate nature of the data must be considered. Log-ratios are a powerful approach for transforming compositional data to an unconstrained vector space of linearly independent coordinates, allowing the application of standard statistical methods. Three of them, the additive, centered, and isometric log-ratios, as well as amalgamations, are introduced in this work, to the chemical engineering community, discussing their potentials, advantages, and possible pitfalls when using them.