An evolutionary K-means algorithm for clustering time series data

It is well known that the K-means clustering algorithm is easy to get stuck at locally optimal points for high dimensional data. Many initialization techniques have been proposed to attack this problem, but with only limited success. We propose an evolutionary K-means algorithm to attack this problem. The proposed algorithm combines genetic algorithms and K-means algorithm together for improving the search ability of the K-means algorithm. We rearrange the clusters in crossover operation based on the distance of clustering centers to avoid generating meaningless offspring. A new genetic operator called swap is proposed to replace the traditional mutation operator for avoiding producing invalid offspring. Experiments performed on some publicly available time series data sets demonstrate the effectiveness and efficiency of the proposed algorithm.