Autoregressive order estimation combined with pruning of the coefficients

R. Moddemeijer

University of Groningen, Department of Computing Science,
P.O. Box 800, NL-9700 AV Groningen, The Netherlands,
phone: +31.50.363 3940 - fax: +31.50.363 38005 - e-mail: rudy@cs.rug.nl

Abstract

A correctly derived Auto Regressive (AR) model can not always optimize the intended approximation. An optimal model should balance bias, caused by under-fitting, and additional variance, caused by over-fitting. The selection of this optimal AR-model is a combination of AR-order estimation and the reduction of the number of coefficients by pruning. We leave the classical approach of AR-order estimation and replace it by AR-coefficient selection. As a selection criterion we use the Modified Information Criterion (MIC), which is closely related to Akaike's criterion (AIC) and has a guaranteed a priori chosen over-fitting probability. We present our algorithm and its verification by simulations.


Full paper


Poster


Published

Signal Processing Symposium (SPS 2000), IEEE Benelux Signal Processing Chapter , March 23-24, 2000, Hilvarenbeek (NL), Note: proceeding are published on CD-rom only, BibTeX
other publications