Accession Number:

ADA175736

Title:

Nearest Neighbor Rule Classification of Time Series in Exploratory Population Screening Problems.

Descriptive Note:

Technical rept.,

Corporate Author:

STANFORD UNIV CA DEPT OF STATISTICS

Personal Author(s):

Report Date:

1986-12-09

Pagination or Media Count:

20.0

Abstract:

In exploratory time series classification problems only a modest number of categorically labeled sample time series is available. The scientific conjecture in this situation is that there is sufficient information in the time series to distinguish between the alternative population categories. The principal objective of the analysis is to determine the separability of the alternative categorical classes, i.e., to estimate the minimum achievable probability of misclassification. A nearest neighbor time series classification rule is advocated for the solution of this problem. In our approach, the dissimilarity between a new-to-be-classified time series and each of the members of the labeled sample time series is computed as an estimate of the Kullback Leibler number, as if the time series were Gaussian distributed. Time and frequency domain formulas for the Kullback Leibler number between stationary multivariate Gaussian time series yield a practical parametric model computational realizatioon of our procedure.

Subject Categories:

  • Statistics and Probability

Distribution Statement:

APPROVED FOR PUBLIC RELEASE