Modify Initialization k-means Clustering Algorithm to Generate Initial Centroids

Abstract

K-means is one of the most common clustering techniques used with numeric data. Different issues are conducted in k-means algorithm in order to reach the optimum solutions with best situations, weather producing good results or the ways used to produce the results efficiently. Initial centroids of this algorithm play important role, so the generation initial centroids attracting more work. However, this paper aims to discuss a new proposed step to improve the generation of initial centroids i.e. modification the first iteration of k-means algorithm. The experiment work of this paper would be applied with one of the famous data that is "iris", this data is suited with k-means algorithm. The experiments were tested with the origin k-means algorithm in two parameters: "execution time" and "cost function" that is represented by sum square error  SSE. The results are promise  work with this modification