The K-L Expansion as an Effective Feature Ordering Technique for Limited Training Sample Size
Marwan Jamil Muasher, D. A. Landgrebe · IEEE Transactions on Geoscience and Remote Sensing · 1983
An effective feature ordering technique is experimentally studied in cases where the number of training samples is limited in classifying multivariate two-class normal distributions. Several experimental results on the Hughes phenomenon using this ordering technique are presented, particularly in situations where the number of training samples is only slightly higher than the number of dimensions.