An improved k-means algorithm for clustering using entropy weighting measures

The objective of traditional k-means algorithm is to make the distances of objects in the same cluster as small as possible, but another objective that the distances of objects from different clusters is not taken into account. This paper presents an improved k-means algorithm satisfying both of objectives above. We modify the cost function of entropy weighting k-means clustering algorithm by adding a variable that is relevant linearly to the square sum of distances from the mean of all objects and the means of all clusters. The improved k-means clustering algorithm is presented and the effectiveness of the algorithm is demonstrated by comparing the results with other k-means clustering algorithms on iris data.