Feature selection via orthogonal expansion of MEG signals

Abstract

The processing of magnetoencephalogram (MEG) signals via orthogonal expansion is examined. The Karhunen-Loeve expansion is used as a tool for feature selection in order to lower the dimensionality of the data and achieve data compression. Data reduction achieved through this method is approximately 6:1. A comparison with the data reduction achieved via autoregressive modeling is made, and the advantages and disadvantages of the Karhunen-Loeve method are discussed.<>