Search

Associative classifiers have been widely used in many domains due to their inherent interpretability. They are built in steps, one of them aimed at ranking the rules, usually performed through objective measures. Works aim to modify this step in order to obtain a classifier with better performance. Among them are those that use multiple measures simultaneously in order to consider different points of view for a given rule. However, these works present problems regarding execution time and interpretability. Here we show the use of ranking aggregation methods, specifically Borda's methods, to rank the rules through a set of measures. Our results demonstrate that our solution is fast to execute and still guarantee the interpretability of the models, since they contain a statistically significant smaller number of rules. [ABSTRACT FROM AUTHOR]

Associative classifiers (ACs) have been widely used in several domains due to their inherent interpretability, with CBA being the most used algorithm in the family. They are built in steps, one of them aimed at ranking the rules, usually performed through objective measures (OMs). Aiming to improve its performance, works seek to modify its sorting step (i) using new measures and/or existing measures individually or (ii) aggregating existing measures to use them simultaneously. However, more than 60 OMs are available in the literature. Therefore, this work aims to group these OMs according to their performance when applied in the ACs sorting step. 15 groups were obtained, which are ranked according to their performance. This work contributes to the improvement and/or development of works, providing guidance for choosing the most appropriate OMs. An example of the impact of the groups obtained is also presented. [ABSTRACT FROM AUTHOR]