Application of evolution strategy in cluster analysis

K-means clustering has two disadvantages, one is easily trapped in local minimum, and the other is difficultly to determine the number of clusters K. To address the problems, this paper proposes 3 new K-means algorithms based on Evolution Strategy. The first individual represents a kind of cluster scheme, and the second represents cluster centers. They can find optimal clustering if K is given. While the third individual adds K on the basis of the first one, it can optimize cluster center and K simultaneously. They all own a simple coding scheme and small population. These algorithms are applied to cluster Fisher´s iris data set and work very well, especially when a priori knowledge is insufficient.