Student:	Miraç MURAT
Supervisor:	Doç. Dr. Şükrü ÖZŞAHİN
Department:	Endüstri Mühendisliği
Institution:	Graduate School of Natural and Applied Sciences
University:	Karadeniz Technical University Turkey

Title of the Thesis:	A NEW LVQ METHOD BASED ON GRAY RELATIONAL ANALYSIS: AN APPLICATION ON CLASSIFICATION OF WOOD SPECIES
Level:	M.Sc.
Acceptance Date:	9/1/2018
Number of Pages:	61
Registration Number:	i3297

Summary:

This thesis concentrates on the development of classification models with artificial

neural network algorithms which are powerful estimation technique. In the thesis, Learning

Vector Quantization (LVQ) one of the artificial neural network techniques is used in addition

to Grey LVQ which is a new LVQ method. The classification performances of LVQ methods

and Multilayer Perceptron (MLP) are compared with an application based on wood type

determination.

The similarities between the sample vector (input vector) and the reference vectors in

the LVQ classification algorithm are determined by evaluating the reference vectors

individually. In the Gray LVQ algorithm, a classification which evaluates all the reference

vectors together using the GRA is performed. To identify 4 different species belonging to

genus of Acer L., classifiers were developed by using the biometric features of wood

anatomy as input of LVQ, Gray LVQ and MLP algorithms. The classification accuracies of

LVQ and Gray LVQ algorithms, which offer a new approach to the rare studies in the

literature in which the wood species are identified using biometric measurements of

anatomical features are similar to the accuracy of the MLP classifier. In all three methods,

95.83% classification accuracy was achieved in the test data set consisting of 24 samples.

Key Words: Classification, Learning Vector Quantization (LVQ), Gray Relational

Analysis, Identification Wood Species