Self-organising adaptive moment-based clustering

Abstract

Known and new strategy elements of clustering methods and of Kohonen's learning mechanism are suitably modified and combined to create a method for self-organising adaptive clustering of time-series data. Each resulting cluster is characterised by a mass, indicating the importance of the cluster and by moment-based parameters, indicating the position and shape of the cluster in the data space. The underlying mechanism updates the cluster parameters to incorporate new data efficiently as it does not require retention of all individual, earlier data points but only a few corresponding aggregated values.