Clustering mixed data using an Artificial Bee Colony

Abstract

In this paper, we have proposed a clustering technique which optimizes the total compactness and separation (measured using the Silhouette index) of the clusters. The proposed algorithm uses an Artificial Bee Colony (ABC) based optimization method as the underlying optimization criterion. We used similarity based prototypes as cluster centers. The proposed clustering technique is able to suitably handle mixed and incomplete data types in such a way that the original characteristics of the data are preserved. Assignment of points to different clusters is done based on a dissimilarity function rather than the Euclidean distance. Results on real-life data sets show that the proposed technique is well-suited to detect true partitioning from data sets. Results are compared with those obtained by four existing clustering techniques, one genetic algorithm based clustering technique (AGKA), the k-Prototypes (KP) algorithm, well-known based K-means clustering technique for similarity functions (KMSF) and a newly developed algorithm with dissimilarity based clustering technique (AD2011).