Statistical Learning for Big Dependent Data

CHAPTER 5CLUSTERING AND CLASSIFICATION OF TIME SERIES

In this chapter, we study how to divide a set of time series into homogeneous groups of series with similar properties and how to classify a time series into one cluster among several possible clusters. The procedures to achieve these objectives are called classification methods. When the classification is made without a reference set, in which similar series are organized in a known number of groups, the procedures are called clustering methods. Such a classification is also called unsupervised classification or unsupervised pattern recognition. When we have series that are classified into groups with labels, and the objective is to classify new observed series, the procedures are called discrimination methods. The classification is then called supervised classification or supervised pattern recognition.

In time series clustering, we have a set of time series for and would like to partition these series into groups (or clusters) such that (1) all series are classified, (2) each series belongs to one and only one group, and ...