Using aggregate objective measures in associative classifiers

Abstract

Associative classification, which has been widely used in several domains, aims to obtain a predictive model in which the process is based on the extraction of association rules (ARs). Model generation occurs in stages, one of them focused on ranking and pruning a set of rules. Regarding ranking, one of the solutions is to sort the rules through objective measures (OMs). Ranking criterion affects the classifier accuracy. This work focus exactly in this stage (ranking). In literature proposals the OMs are explored separately. However, there are works, in the ARs domain, that explore the aggregate use of OMs. Based on these ideas, [Silva and Carvalho, 2018] explored measures aggregation, in which several OMs are considered at the same time, in associative classification context. [Silva and Carvalho, 2018] used the aggregation strategy proposed by [Bouker et al., 2014]. Although [Bouker et al., 2014] had used the concept of non-dominance to rank the rules, the authors leave open the possibility of ranking by the concept of dominance. This dual concept of dominance and non-dominance is also explored in [Dahbi et al., 2016]. Considering the above, this project aims to: (a) jointly explore the concepts of dominance and non-dominance to rank the rules; (b) explore the aggregation of measures considering another perspective, that of an ensemble of classifiers.