Clustering methods for the identification of structured composite sources

Results are presented concerning the problem of identifying temporal structure in composite sources. An alternative class of techniques for identifying the underlying structure of a SCS (structured composite source) from its estimated transition matrix is proposed. These techniques are postulated directly with respect to the discrete elements of a Markov chain and allow for non-hierarchical and hierarchical decomposition. The general structure of this class is developed, and examples based on a specific clustering algorithm are discussed.<
>