A method of two stage clustering using agglomerative hierarchical algorithms with one-pass k-means++ or k-median++

The aim of this paper is to propose a two-stage method of clustering in which the first stage uses one-pass k-median++ and the second stage uses an agglomerative hierarchical clustering. To handle medians in the second stage, we proposed two calculation methods. One method uses L₁ distance as similarity. Another uses error of L₁ distance like the Ward method. In this paper, we compared proposed method and a two-stage method of our study which uses k-means++ in the first stage to examine the effectiveness of L₁ distance in two-stage methods. Numerical experiments have been done using two criteria: objective function values and the Rand index.