Enhancing the Performance of Evolutionary Clustering by Genetic Sequence Resorting

As a global optimization technique, the evolutionary algorithms provide powerful solvers to the data clustering problem. However, the individual representations in evolutionary clustering algorithms exist redundancy and inconsistency problems, which not only wastes the search efforts but also disorders the mutual learning process between individuals. To address the problems, in this paper, we propose a genetic sequence resorting method. This method first identifies a reference parameter vector and then resorts the order of the genetic sequences in each individual according to their distance to the reference vector. In this way, the solutions represented by individuals are unified, which removes redundancy and enhances the efficiency of the mutual learning in the algorithm. Incorporating the above method, a new and generic evolutionary clustering framework is developed. Under this framework, we specifically design two algorithms: one for distance-based convex clustering and the other for density-based nonconvex clustering. Experiments show that our method can effectively improve the performance of evolutionary clustering algorithms on various datasets with both convex and nonconvex cluster shapes.