k-MM: A Hybrid Clustering Algorithm Based on k-Means and k-Medoids

k-means and k-medoids have been the most popular clustering algorithms based on partitioning for many decades. When using heuristics such as Lloyd's algorithm, k-means is easy to implement and can be applied on large data sets. However, it presents drawbacks like the inefficiency of the used metric, the difficulty of the choice of the input k and the premature convergence. In contrast, k-medoids takes more time to come up with a clustering but ensures a better quality of the result. Moreover, it is more robust to noise and outliers. In this article, we design a hybrid algorithm, namely k-MM to take advantage of both algorithms. We experimented k-MM and we show that, when compared to k-means and k-medoids, it is very efficient and effective. We present also an application to image clustering and show that k-MM has the ability to discover clusters faster and more effectively than a recent work of the literature.