Analysis of Locality in Hybrid Evolutionary Cluster Optimization

State of the art algorithms for cluster geometry optimization rely on hybrid approaches that combine the global exploration performed by evolutionary methods with local search procedures. These methods use derivative information to discover the nearest local optimum. In this paper we analyze the locality properties of this approach to gain insight on the algorithm’s strengths and weaknesses and to determine the role played by each of its components. Results show that there are important differences in what concerns the locality of different mutation operators commonly used in this problem.