Neurocomputing 51 (2003) 87–103 www.elsevier.com/locate/neucom On-line pattern analysis by evolving self-organizing maps Da Deng ∗ , Nikola Kasabov Department of Information Science, University of Otago, Dunedin, New Zealand Received 13 July 2001; accepted 13 February 2002 Abstract Many real world data processing tasks demand intelligent computational models with good eciency and adaptability in their on-line operations. Consequently, neural algorithms with con- structive network structure and incremental learning ability are of increasing interest. In this paper we present an algorithm of evolving self-organizing map (ESOM), which features an evolving network structure and fast on-line learning. Experiments have been carried out on some bench- mark data sets for vector quantisation and classication tasks. Compared with other methods, ESOM achieved better or comparable performance with a much shorter learning process. Our results show that ESOM is a promising computational model for on-line pattern analysis in real world problems. c  2002 Elsevier Science B.V. All rights reserved. Keywords: On-line learning; Self-organizing; Clustering; Classication 1. Introduction In this explosive era of information growth many real world information systems need to process on-line data streams that are updated frequently, such as stock market indexes and video streams transferred across the Internet. To manipulate the large amount of on-line data and extract useful information, a number of techniques such as visualisation, clustering and classication, etc. need to be explored. A few diculties, however, exist here, such as • Input and output distributions of data are not known a priori and these distributions may change over time. * Corresponding author. E-mail addresses: ddeng@infoscience.otago.ac.nz (D. Deng), nkasabov@infoscience.otago.ac.nz (N. Kasabov). 0925-2312/03/$-see front matter c  2002 Elsevier Science B.V. All rights reserved. PII:S0925-2312(02)00599-4