Histogram-Based Method for Effective Initialization of the K-Means Clustering Algorithm

K-means is undoubtedly the most widely used partitional clustering algorithm. Unfortunately, this algorithm is highly sensitive to the initial selection of the cluster centers. Numerous initialization methods have been proposed to address this drawback. Many of these methods, however, have superlinear complexity in the number of data points, which makes them impractical for large data sets. On the other hand, linear methods are often random and/or sensitive to the order in which the data points are processed. These methods are generally unreliable in that the quality of their results is unpredictable. In this paper, we propose a linear, deterministic, and order-invariant initialization method based on multidimensional histograms. Experiments on a diverse collection of data sets from the UCI Machine Learning Repository demonstrate the superiority of our method over the well-known maximin method.