An initial centroid selection method based on radial and angular coordinates for K-means algorithm

Clustering is a technique for dividing a set of similar objects into same groups and dissimilar objects into different groups. Among different clustering algorithms, the K-means algorithm is considered as the most popular due to its simplicity. However, the outcome from the K-means algorithm is highly sensitive to the initial centroid selection. As a consequence, the selection of initial centroids in the K-means algorithm plays a crucial part in accuracy and efficiency. To select the initial centroids more effectively, in this paper, we propose a new method based on radial and angular coordinates. To check the feasibility of the proposed method, we compare our method with the standard K-means algorithm. For the comparison, we use synthetic data sets with different size of instances and number of clusters. The experiment shows that in most of the cases the proposed method clearly dominates over the standard K-means algorithm in terms of execution time and required number of iterations.