Automatic clustering using particle swarm optimization with various validity indices

Abstract

Data clustering partitions a dataset into clusters where each cluster contains similar data. Clustering algorithms usually require users to set the number of clusters, e.g., k-means or fuzzy c-means. However, it is difficult to determine a meaningful number of clusters if users lack prior knowledge of the data. Data clustering may use a validity index to grade the clustering quality. Most validity indices are based on clustering compactness and separation, but other criteria are also used for clustering. Therefore, no individual validity index is applicable to data with different properties. This paper presents a novel dynamic clustering based on particle swarm optimization. The proposed algorithm is compared with other dynamic clustering algorithms based on particle swarm optimization using artificial and real data sets. The experimental results showed that our proposed algorithm not only determines the appropriate number of clusters with correct cluster centers but can also be applied to data with different properties using various validity indices.