Publication details

Recent feature selection methods in statistical pattern recognition

Monography Chapter

Pudil Pavel, Novovičová Jana, Somol Petr

serial: Pattern Recognition and String Matching, p. 1-51 , Eds: Chen D., Cheng X.

publisher: Kluwer, (Dordrecht 2003)

research: CEZ:AV0Z1075907

project(s): GA402/01/0981, GA ČR, KSK1019101, GA AV ČR

keywords: pattern recognition, feature selection, search methods

abstract (eng):

The chapter is devoted to the problem of feature selection in statistical pattern recognition. A number of feature subset search strategies is reviewed, analyzed and compared. New algorithms are described (Fast Branch and Bound, Branch and Bound Algorithm with Partial Prediction, Floating Search, Adaptive Floating Search). Two feature selection methods based on approximating the unknown class conditional densities by finite mixtures of the factorized densities are presented.

Cosati: 09K, 12B