ABSTRACT
This chapter addresses the case of multivariate time-series clustering and utilizes spatiotemporal data as specific application domain of interest. Time-series data is one of the most common forms of data encountered in a wide variety of scenarios such as the stock markets, sensor data, fault monitoring, machine state monitoring, environmental applications, or medical data. Time-series data allow diverse formulations for the clustering process, depending upon whether the series are clustered on the basis of their online correlations, or whether they are clustered on the basis of their shapes. Online correlation-based clustering methods are closely related to the problem of forecasting. Such methods are typically based on clustering the streams on the basis of their correlations with one another in their past window of history. Sensor selection algorithms are naturally related to correlation clustering. This is because such methods typically pick a representative set of streams which can be used in order to predict the other streams in the data.
