Classification of Long Memory Processes: A Simulation Study

• Activity Number: 637
• Type: Contributed
• Date/Time: Thursday, August 5, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Ritaja Sur*+ and Benjamin Kedem
• Companies: University of Maryland and University of Maryland
• Keywords: Time Series ; Classification ; Cluster analysis ; Distance measure ; Long memory ; FEXP model

Abstract

The study of time series classification using clustering techniques has been an area of great interest in various fields. For this purpose of classification, there is a need to identify similarities or dissimilarities in the data. This similarity can be computed using a time or frequency domain measure. One such measure computed in the frequency domain is based on the expansion of the log spectral density of the stationary time series in terms of a cosine basis. This measure has shown good results for both linear and non-linear time series data. We look at the application of this measure in clustering to various long memory processes. In addition, we study the classification results with the measure obtained from the more generalized model known as fractional exponential model. Simulation studies comparing the two measures for various long memory processes will be presented.