Abstract: After optimizing random selection process of the initial cluster centers,an anomaly detection algorithm based on improved K-means clustering was proposed.When the cluster centers are selected,the tightness of all data points is calculated, outliers region is removed,and then the K initial centers in dense regions of data are selected,which avoids that the random selection is easy to cause the defect of local optimum.By optimizing the selection process,the initial cluster centers are more closer to the real clusters centers before iteration of the algorithm,the numbers of iterations are reduced,and the quality of clustering and anomaly detection rate are improved.Experiments show that the improved algorithm is much better than the original algorithm in clustering performance and anomaly detection.

Key words: K-mean,Clustering,Tightness,Anomaly detection