Date of Award

Spring 2001

Document Type

Thesis - Restricted

Degree Name

Master of Science (MS)

Department

Electrical and Computer Engineering

First Advisor

Feng, Xin

Second Advisor

Povinelli, Richard

Third Advisor

Johnson, Michael

Abstract

This thesis proposes an improved time series temporal pattern identification method, which focuses on finding the temporal pattern clusters in time series. Based on a new concept of the fuzzy temporal pattern cluster, a new differentiable fuzzy cluster objective function is generated. Then a two step optimization algorithm is presented to find the optimal temporal pattern clusters in time delay embedding phase space, and greatly improves the time performance of the optimization procedure. The new method is distinguished from previous time series analysis systems by its fuzzy temporal pattern clustering method. The system's advantages over previous efforts include the efficiency and consistency of the system algorithms. Brief overviews of the related field of non-linear time series processing, optimization and previous work in time series data mining (TSDM) are included. The time series temporal pattern identification method is presented in details, followed by applications of the system to several time series analysis problems and algorithm comparisons with two other systems. At last, a real software system is designed to facilitate the time series data mining research. This system also presents and maintains a financial time series database, which could be applied in the future research.

Share

COinS

Restricted Access Item

Having trouble?