Repositori URV

Identificador: imarina:9002813

Handle: https://hdl.handle.net/20.500.11797/imarina9002813

Autors:
Santana PJLanzarini LBariviera AF

Resum:
© 2020 J. Adv. Inf. Technol. Having strategies capable of automatically generating classification rules is highly useful in any decision-making process. In this article, we propose a method that can operate on nominal and numeric attributes to obtain fuzzy classification rules by combining a competitive neural network with an optimization technique based on variable population particle swarms. The fitness function that controls swarm movement uses a voting criterion that weights, in a fuzzy manner, numeric attribute participation. The efficiency and efficacy of this method are strongly conditioned by how membership functions to each of the fuzzy sets are established. In previous works, this was done by partitioning the range of each numeric attribute at equal-length intervals, centering a triangular function with appropriate overlap in each of them. In this case, an improvement to the fuzzy set generation process is proposed using the Fuzzy C-Means methods. The results obtained were compared to those yielded by the previous version using 11 databases from the UCI repository and three databases from the Ecuadorian financial system – one from a credit and savings cooperative and two from banks that grant productive and non-productive credits as well as microcredits. The results obtained were satisfactory. At the end of the article, our conclusions are discussed and future research lines are suggested.