MVAR-Geo: A Proposal for Visualizing Georeferenced Association Rules

Abstract

The rapid development of geographic localization techniques has provided the means for generating and increasingly collecting massive georeferenced datasets. Geographic Information Systems (GIS) have been used to process and represent this kind of data visually on maps in the search for a better understanding of the possible relationships between geographic areas and the variables of the represented application domain. The objective of its use is to reduce the time it takes to analyze information, in addition to enabling access to spatial patterns that would be difficult to identify with simplified data listings or graphs (e.g., bars, lines, etc.) that are not very informative in a spatial context. Some modern versions of GIS have sought the support of data mining, among which Association Rule Mining (ARM) is a good option due to its relative simplicity and transparency. The combination of GIS and ARM can support the development of more sophisticated analyses, improve spatial analysis, and enable the identification of patterns of interest and the extraction of valuable knowledge. However, there are still a few works in the literature that propose this type of solution, not exploring the full potential that this approach could offer to visually communicate patterns and trends that are difficult to identify through conventional data analysis methods, such as graphs and tables. To address this gap, the objective of this work is to introduce MVAR-Geo, a method for visualizing association rules in georeferenced data. To demonstrate how this proposal can benefit the analysis of association rules with georeferenced attributes, we apply the method in three case studies to conduct spatial analyses of data from different application domains using levels of geographic granularity in accordance with each context analyzed. The case studies demonstrated that MVAR-Geo significantly enhances the ability to interpret spatial data by enabling the integration of ARM with GIS. The use of interest measures from association rules in geographic visualizations, particularly the lift measure, has proven essential for distinguishing genuinely strong associations from those expected by chance, improving the understanding of how specific geographic areas can influence the chances of occurrence of certain events. The proposed method also favors identifying spatial patterns involving agglomerations, anomalies, and geographic subspaces , which corroborates the potential contribution of this work.