Partitioning Techniques
Lynne Billard, Edwin Diday · 2019
This chapter explains how the partitions are obtained for symbolic data. Partitioning methodology is perhaps the most developed of all clustering techniques, at least for classical data, with many different approaches presented in the literature, starting from the initial and simplest approach based on the coordinate data by using variants of the k-means methods, or procedures based on dissimilarity/distance measures with the use of the k-medoids method, or their variations. This includes the more general dynamical partitioning of which the k-means and k-medoids methods are but special cases. The chapter considers multi-valued list observations and interval-valued observations, followed by histogram observations. In most methods developed thus far for symbolic data, it is assumed that all variables have the same type of observation. The chapter also considers an example of mixed-valued observations, and examines mixture distribution models. It provides a short description of issues involved in calculating the final cluster representations.