Change Detection in Rough Time Series
Abstract
A discrete time series may characteristically have high noise levels resulting in a rough or jagged distribution which can present significant challenges to conventional statistical tracking techniques. To address this problem the proposed method applies hybrid fuzzy statistical techniques to series granules instead of to individual measures. After detailing the method and its rationale, three examples demonstrate the robust nature of the proposed fuzzy tracking signal which leads to a minimal number of false alarms caused by isolated spikes. The examples demonstrate the effectiveness of this tracking signal for promptly identifying significant pattern changes in rough time series as can be encountered in data sets used for various types of Defence decision making.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2014
- Accession Number
- ADA615321
Entities
People
- Lewis Warren
Organizations
- Defence Science and Technology Group