Towards a Fast Evolutionary Algorithm for Clustering

This paper elaborates on the improvement of an evolutionary algorithm for clustering (EAC) introduced in previous work. Four new features are proposed and empirically assessed in seven datasets, using two fitness functions. Statistical analyses allow concluding that two proposed features lead to significant improvements on the original EAC. Such features have been incorporated into the EAC, resulting in a more computationally efficient algorithm called F-EAC (Fast EAC). We describe as an additional contribution a methodology for evaluating evolutionary algorithms for clustering in such a way that the influence of the fitness function is lessened in the assessment process, what yields analyses specially focused on the evolutionary operators.