Feature selection Evaluation, application, and small sample performance.pdf

Feature Selection Evaluation Application and Small Sample Performance Anil Jain Douglas Zongker Department of Computer Science Department of Computer Science Michigan State University Michigan State University East Lansing Michigan USA East Lansing Michigan USA fjain ;zongker g cpsmsu edu Abstract A large numb er of algorithms have b een prop osed for feature subset selection Our exp erimental results show that the sequential forward oating selection SFFS algorithm prop osed by Pudil et al dominates the other algorithms tested We study the problem of cho osing an optimal feature set for land use classication based on SAR satellite images using four dierent texture mo dels Po oling features derived from dierent texture mo dels followed by a feature selection results in a substantial improvement in the classication accuracy We also illustrate the dangers of using feature selection in small sample size situations Keywords feature selection curse of dimensionality genetic algorithm no de prun ing texture mo dels SAR image classication 1 Intro duction The problem of feature selection is dened as follows given a set of candidate features select a subset that p erforms the b est under some classication system This pro cedure can reduce not only the cost of recognition by reducing the numb er of features that need to b e collected but in some cases it can also provide a b etter classication accuracy due to nite sample size eects The term fe ature selection is taken to refer to algorithms that output a subset of the input feature set More general