Many objective evaluation measures (OMs) have been proposed in the last years as a mean of post-processing association rules. Therefore, the first challenge during an exploration process is to decide which OM to use. For that, one can: (a) reduce the number of OMs to be chosen; (b) aggregate OMs' values in one importance's value as a mean of not selecting a suitable OM. The problem with (a) is that many OMs can remain. Regarding (b), an optimal equation, that derives a single value, is generally obtained, which cannot be well understandable by the user. In this context, [Carvalho et al., 2015] propose a process to cluster association rules, based on the existing similarity among the rankings obtained by a set of OMs, in order to direct the user to the interesting patterns of the domain. The idea is to solve the problem related to the identification of a set of suitable OMs, by implicit combining OMs, without using any optimization method. Therefore, with this process (I) it is not necessary to select a set of suitable OMs nor explicit aggregate many OMs, in order to rank the rules to find the interesting ones; (II) the exploration space can be reduced since it is considered that there is a subset of groups that contains all the interesting rules, so that a small number of groups have to be explored. However, the described process presents some gaps to be explored.Based on the exposed, this project aims to improve the process proposed by [Carvalho et al., 2015] in order to: (a) explore ways to rank the clusters so that the user can explore only the n first groups (the ones that will contain the interesting patterns); (b) explore alternative ways to rank the rules inside the clusters to try to improve the results; (c) explore the results of the process when the OMs used in the clustering are redundant (i.e., lead to the same ranking).
News published in Agência FAPESP Newsletter about the scholarship: