Marginality: A Numerical Mapping for Enhanced Exploitation of Taxonomic Attributes

Hierarchical attributes appear in taxonomic or ontologybased data (e.g. NACE economic activities, ICD-classified diseases, animal/plant species, etc.). Such taxonomic data are often exploited as if they were flat nominal data without hierarchy, which implies losing substantial information and analytical power. We introduce marginality, a numerical mapping for taxonomic data that allows using on those data many of the algorithms and analytical techniques designed for numerical data. We show how to compute descriptive statistics like the mean, the variance and the covariance on marginality-mapped data. Also, we define a mathematical distance between records including hierarchical attributes that is based on marginality-based variances. Such a distance paves the way to re-using on taxonomic data clustering and anonymization techniques designed for numerical data.