An Extended Version of the Gustafson‐Kessel Algorithm for Evolving Data Stream Clustering
Dimitar Filev, Olga Georgieva · 2010
This chapter presents one approach to the problem of discovering structures of streaming data in real time. The proposed evolving Gustafson-Kessel-like (eGKL) clustering algorithm is a step toward the problem of online evolving clustering of streams of data. It provides a methodology for adaptive, step-by-step identification of clusters that are similar to the Gustafson-Kessel (GK) clusters. The algorithm is based on a new approach to the problem of estimation of the number of clusters and their boundaries that is motivated by the identified similarities between the concept of evolving clustering and the well-established multivariate statistical process control (SPC) techniques. The proposed algorithm addresses some of the problems of evolving systems, for example, recursive estimation of the cluster centers, inverse covariance matrixes, covariance determinants, Mahalanobis distance-type similarity relations that are enablers for computationally efficient implementation in real-time identification, classification, prognostics, process modeling, anomaly detection, and other embedded-system types of applications. Controlled Vocabulary Terms process control; recursive estimation; statistical process control; streaming media; workstation clusters