A Survey of Recent Methods for Efficient Retrieval of Similar Time Sequences

Abstract

Time sequences occur in many applications, ranging from science and technology to business and entertainment. In many of these applications, an analysis of time series data, and searching through large, unstructured databases based on sample sequences, is often desirable. Such similarity-based retrieval has attracted a lot of attention in recent years. Although several different approaches have appeared, most are based on the common premise of dimensionality reduction and spatial access methods. This paper gives an overview of recent research and shows how the methods fit into a general context of signature extraction.

The Basic Principles of Metric Indexing

“…One example of this technique is similarity indexing for time series, where many signature types have been suggested, normally in the form of fixeddimensional vectors that may be indexed using spatial access methods. Hetland [100] gives a survey of this application. 9.…”

“…For a discussion of how to deal with high-dimensional vector spaces, see the survey by Hetland [100]. It is interesting to note that the intrinsic dimensionality of a vector data set may be lower than its representational dimensionality; that is, the vectors may be mapped faithfully to a lower-dimensional vector space without distorting the distances much.…”

Evolutionary Rule Mining in Time Series Databases

“…Other usable methods of discretization include those used to simplify time series for indexing purposes. See Hetland (2004) for a survey. A recent discretization method with several interesting properties is the SAX method (Lin et al, 2003).…”

Case Based Representation and Retrieval with Time Dependent Features

“…In the literature, most of the approaches to similarity-based time series retrieval are founded on the common premise of dimensionality reduction (see the survey in [12]). As a matter of fact, a discretized time series can always be seen as vector in an n-dimensional space (with n typically extremely large).…”

“…One obvious solution is thus to reduce the time series dimensionality, by means of a transform that preserves the distance between two time series, or underestimates it: in this case a post-processing step will be required, to filter out the so-called "false alarms"; the requirement is never to overestimate the distance, so that no "false dismissals" can exist [12]. Widely used transforms are the Discrete Fourier Transform (DFT) [2], and the Discrete Wavelet Transform (DWT) [9].…”