Measure based metrics for aggregated data

Article type: Research Article

Authors: Rayward-Smith, V.J.

Affiliations: School of Computing Sciences, University of East Anglia, Norwich, NR4 7TJ, UK. E-mail: [email protected]

Abstract: Aggregated data arises commonly from surveys and censuses where groups of individuals are studied as coherent entities. The aggregated data can take many forms including sets, intervals, distributions and histograms. The data analyst needs to measure the similarity between such aggregated data items and a range of metrics are reported in the literature to achieve this (e.g. the Jaccard metric for sets and the Wasserstein metric for histograms). In this paper, a unifying theory based on measure theory is developed that establishes not only that known metrics are essentially similar but also suggests new metrics.

Keywords: Measure theory, metric space, similarity, clustering, symbolic data, aggregated data

DOI: 10.3233/IDA-2010-0459

Journal: Intelligent Data Analysis, vol. 15, no. 2, pp. 109-130, 2011