Ferramentas para contextualização geográfica de outliers em conjuntos de dados multidimensionais

Abstract

Analyzing, contextualizing, and understanding outliers in complex datasets, with many heterogeneous attributes, presents big challenges. For the specialist performing the analysis, it is not always trivial to identify which attributes are relevant to the problem at hand, even with the usage of data visualization techniques. This problem is even more challenging in datasets that demand the geographic interpretation of outliers, such as (i) meteorological data; (ii) demographic census data; (iii) socio-economic data from several cities. The present work proposes tools for simplifying the task of geographic contextualization and interpretation of outliers, through visualizations generated with the help of Outlying Aspect Mining algorithms. With these tools, it is expected that more accurate, direct, and efficient analyses are possible, allowing the specialist to understand and contextualize outliers more easily, from a geographic perspective. As a test case, public data on vaccination against Covid-19 in Brazil, made available by OpenDataSus, will be used.