Summary: Comparison of Statistical and Neural Classifiers
and Their Applications to
Optical Character Recognition and Speech Classification
Ethem Alpaydin, Fikret G¨urgen
Department of Computer Engineering
Bo–gazi¸ci University
TR­80815 — Istanbul Turkey
falpaydin,gurgeng@boun.edu.tr
Neural Network Systems Techniques and Applications (in print)
C. T. Leondes (Ed.), c
flACADEMIC Press
October 24, 1996
Abstract
We give a review of basic statistical and neural techniques for classification. Statistical tech­
niques are based on the idea of estimating class­conditional likelihoods and using Bayes rule
to convert these to posterior class probabilities whereas neural techniques estimate directly the
posteriors. Statistical techniques include (i) Parametric (Gaussian) Bayes classifiers, (ii) Non­
parametric kernel­based density estimators like k­nearest neighbor and Parzen windows, and
(iii) mixtures of (Gaussian) densities (a special case of which is the Learning Vector Quanti­
zation). As neural classifiers, we include simple perceptrons and multilayer perceptrons with

Source: Alpaydýn, Ethem - Department of Computer Engineering, Bogaziçi University

Collections: Computer Technologies and Information Sciences