Detecting groups of anomalously similar objects in large data sets

Issue title: Selected papers from IDA2005, Madrid, Spain

Article type: Research Article

Authors: Zhang, Zhicheng^{; *} | Hand, David J.

Affiliations: Department of Mathematics, Imperial College London, 180 Queens Gate, Huxley Building, London, SW7 2AZ, UK. E-mail: zhzhang@imperial.ac.uk, d.j.hand@imperial.ac.uk

Note: [*] The work of the first author was supported by an EPSRC CASE award in conjunction with GlaxoSmithKline. We are particularly grateful for the support given by Dr Steven Barrett.

Abstract: Pattern discovery is a facet of data mining concerned with the detection of “small local” structures in large data sets. In high dimensions this is typically difficult because of the computational work involved in searching over the data space. In this paper we outline a tool called PEAKER which can detect patterns efficiently in high dimensions. We approach the subject through the two aspects of pattern discovery, detection and verification. We demonstrate various ways of using PEAKER as well as its various inherent properties, emphasizing the exploratory nature of the tool.

Keywords: Pattern discovery, pattern detection, pattern verification, PEAKER

DOI: 10.3233/IDA-2006-10506

Journal: Intelligent Data Analysis, vol. 10, no. 5, pp. 473-483, 2006