RESOLUÇÃO DO PROBLEMA DOS K-MEDOIDS VIA ALGORITMO GENÉTICO DE CHAVES ALEATÓRIAS VICIADAS

This paper proposes a new Optimization Algorithm for the k-medoid Clustering Problem.In this problem, given a dataset X with n objects and f attributes and a fixed number of clusters (k), it is necessary to select k objects called medoids.Each medoid creates a new cluster and the remaining (n-k) objects should be placed into nearest of these clusters, according a distance measure.The goal is minimize the sum of distances between each object and the medoid of its group.This work presents a new algorithm that considers concepts of Biased Random-key Genetic Algorithm.Besides, an approach of path-relinking procedure is related.The computational experiments results are presented in the last section.It was used thirty instances, among well-known datasets of the literature and new datasets artificially constructed.The proposed heuristics are compared with five approaches (four algorithms and one exact method) and the presented algorithms are a alternative and effective way to solve the problem.