This paper focuses on the discovery of surprising, unexpected patterns, based on a data mining method that consists of detecting instances of Simpson's paradox. By its very nature, instances of this Paradox tend to be surprising to the user. Previous work in the literature has proposed an algorithm for discovering instances of that paradox, but it addressed only "flat" data stored in a single relation. This work proposes a novel algorithm that considerably extends that previous work, by discovering instances of Simpson's paradox in hierarchical multidimensional data
Carem C. Fabris, Alex Alves Freitas