Measurement, selection, and visualization of association rules: A compositional data perspective: A Compositional Data perspective on Association Rules

Abstract

Association rule mining is a powerful data analytic technique used for extracting information from transaction databases with a collection of itemsets. The aim is to indicate what item goes with what item (ie, an association rule) in a set of collected transactions. It is extensively used in text analytics of text records or social media. Here we use Compositional Data analysis (CoDa) techniques to generate new visualizations and insights from association rule mining. These CoDa methods show the relationship between itemsets, their strength, and direction of dependency. Moreover, after expressing each association rule as a contingency table, we discuss two statistical tests to guide identification of the relevant rules by analyzing the relative importance of the elements of the table. As an example, we use these visualizations and statistical tests for investigating the association of negative mood emotions to various types of headache/migraine events. Data for those analysis comes from N1-HeadacheTM, a digital platform where individual users record attacks and symptoms as well as their daily exposure to a list of potential factors.