FC-Kmeans: Fixed-centered K-means algorithm

Abstract

Clustering is one of the data mining methods that partition large-sized data into subgroups according to their similarities. K-means clustering algorithm works well in spherical or convex data distribution of large-sized data sets. Most of the algorithms based on K-means have generally been interested in an initial cluster centers selection or cluster distribution. However, these algorithms may not meet satisfy some requirements in practice. This paper presents the FC-Kmeans algorithm, which enables clustering by fixing some cluster centers considering real conditions. Thus, while some of the cluster centers are fixed, it is tried to obtain the most appropriate cluster centers for the others and the best distribution of the data to the clusters. The K-means clustering algorithm is compared with two different fixed-centered clustering algorithms which are FC-Kmeans and FC-Kmeans 2. The experimental results show that although the FC-Kmeans algorithm has more limitations than K-means, it converges the performance of K-means algorithm according to some performance indicators such as SSE, DB Index and Silhouette Index.