IJPAM: Volume 81, No. 5 (2012)


Natthinee Deetae$^1$, Saowanit Sukparungsee$^2$
Yupaporn Areepong$^3$, Katechan Jampachaisri$^4$
$^{1,2,3}$Department of Applied Statistics
Faculty of Applied Science
King Mongkut's University of Technology
North Bangkok 10800, THAILAND
$^4$Department of Mathematics
Faculty of Science
Naresuan University
Phitsanulok, 65000, THAILAND

Abstract. Classification is emphasized on allocating new observations in the test set of sample to labeled classes based on constructed rule from the training set. With the hybrid of several classification techniques has been developed and mostly exhibited results superior to a single classification technique. The aim of this study is to develop a new classification technique using Empirical Bayes in combination with Nearest Neighbor (EBNN) in the case of unknown mean and known variance. The realization of estimated hyper-parameters obtained from Empirical Bayes (EB) were adjusted using Nearest Neighbor method (NN), providing improved prediction of class membership when compared to that using single method. Data employed in this study are generated, consisting of training set and test set with the sample sizes 100, 200 and 500 for the binary classification. The results indicated EBNN method exhibited an improved performance over EB method in all situations under study.

Received: September 27, 2012

AMS Subject Classification: 62H30, 62F15, 62C12

Key Words and Phrases: classification, empirical bayes, nearest neighbor, posterior predictive probability, Markov chain Monte Carlo (MCMC)

Download paper from here.

Source: International Journal of Pure and Applied Mathematics
ISSN printed version: 1311-8080
ISSN on-line version: 1314-3395
Year: 2012
Volume: 81
Issue: 5