GENETIC-ANT CLUSTERING ALGORITHMS

Abstract

This paper proposed a hybrid clustering algorithm, namely genetic-ant clustering algorithm (GACA), by hybridizing genetic algorithms (GAs) and ant colony system (ACS). The core logic of the proposed algorithm is to employ GAs to optimally determine cluster seeds and then use ACS to optimally assign the remaining objects to one of the cluster seeds. To compare the performance of proposed hybrid algorithm with other algorithms, a numerical study on various scales of clustering problems, including 50, 100 and 300 randomly generated two-dimension objects and a case study on a p-median problem containing 316 cities are conducted, respectively. Moreover, to examine the performance difference between the hierarchical and nonhierarchical clustering methods, additional capacity constraints are imposed. The results of the uncapacitated clustering example and p-median problem consistently show that, in term of effectiveness, our previously proposed hierarchical clustering method -- GCA-CSPM significantly performs best, followed by GACA. However, if the capacity constraint is further introduced, GACA will perform best, implying that GACA is suitable for solving more complex clustering problem which exists highly interacted relationships among objects.