Pattern Clustering using Ants Colony, Ward Method and Kohonen Maps

The goal of this paper is to propose improvements to the ACA (Ant-based Clustering Algorithm), and evaluate its performance relative to the Ward Method; to the One-dimensional Kohonen Maps and to the ACAM (Ant-based Clustering Algorithm Modified) algorithm. The algorithm containing the improvements will be referred here by “proposed” algorithm. Its the main changes were: the introduction of a comparison between the probability of dropping a pattern at the position chosen randomly and the probability of dropping this pattern at its current position; the introduction of an evaluation of the probability of a neighboring position when the decision to drop a pattern is positive and the cell in which the pattern should be dropped is occupied; and the replacement of the pattern carried by an ant, in case this pattern is not dropped within 100 consecutive iterations. To assess the performance of the proposed algorithm three real and public databases were used (Iris, Wine and Pima Indians Diabetes). The results showed superiority of the proposed algorithm when comparing with the ACAM algorithm in two of the three databases.