Using Graph Convolutional Networks to Rank Rules in Associative Classifiers

Maicon Dall’Agnol, Veronica Oliveira de Carvalho, Daniel Pedronette

2025

Abstract

Associative classifiers are a class of algorithms that have been used in diverse domains due to their inherent interpretability. For models to be induced, a sequence of steps is necessary, one of which is aimed at ranking a set of rules. This sorting usually occurs through objective measures, more specifically through confidence and support. However, as many measures exist, new ranking methods have emerged with the aim of (i) using a set of them simultaneously, so that each measure can contribute to identify the most important rules and (ii) inducing models that present a good balance between performance and interpretability in relation to some baseline. This work also presents a method for ranking rules considering the same goals ((i);(ii)). This new method, named AC.RANKGCN, is based on ideas from previous works to improve the results obtained so far. To this end, ranking is performed using a graph convolutional network in a semi-supervised approach and, thus, the importance of a rule is evaluated not only in relation to the values of its OMs, but also in relation to its neighboring rules (neighborhood) considering the network topology and a set of features. The results demonstrate that AC.RANKGCN outperforms previous results.

Paper Citation

in Harvard Style

Dall’Agnol M., Oliveira de Carvalho V. and Pedronette D. (2025). Using Graph Convolutional Networks to Rank Rules in Associative Classifiers. In Proceedings of the 27th International Conference on Enterprise Information Systems - Volume 1: ICEIS; ISBN 978-989-758-749-8, SciTePress, pages 317-325. DOI: 10.5220/0013364700003929

in Bibtex Style

@conference{iceis25,
author={Maicon Dall’Agnol and Veronica Oliveira de Carvalho and Daniel Pedronette},
title={Using Graph Convolutional Networks to Rank Rules in Associative Classifiers},
booktitle={Proceedings of the 27th International Conference on Enterprise Information Systems - Volume 1: ICEIS},
year={2025},
pages={317-325},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013364700003929},
isbn={978-989-758-749-8},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 27th International Conference on Enterprise Information Systems - Volume 1: ICEIS
TI - Using Graph Convolutional Networks to Rank Rules in Associative Classifiers
SN - 978-989-758-749-8
AU - Dall’Agnol M.
AU - Oliveira de Carvalho V.
AU - Pedronette D.
PY - 2025
SP - 317
EP - 325
DO - 10.5220/0013364700003929
PB - SciTePress