UDC 005.519.8
KOCHONEN CARDS FOR MIXED DATA
A. A. Godeaev, a student of the group 343; This email address is being protected from spambots. You need JavaScript enabled to view it.
A. V. Gigolaev, a student of the group 443; This email address is being protected from spambots. You need JavaScript enabled to view it.
N. I. Tsukanova, Ph.D., Associate Professor of the Department of the VPM RSREU; This email address is being protected from spambots. You need JavaScript enabled to view it.
Problems of creation self-organizing Kohonen maps for mixed data and ways of their resolution are examined. It is proposed to select a metric from an allowed set of metrics for each data set and to use fuzzy sets to represent the weights of neurons. The aim of the paper is to create a program which will create selforganizing Kohonen maps and be capable of processing mixed data types using different metrics and to research relations between chosen metric and input data set features with this program. Following problems are discussed: choices of a metric for every data type, data structures for categorical attributes of neuron’s weight vector, the algorithm of adjusting these weights to the input effect and methods of dead neurons elimination. Developed program is described and research results obtained with its help are presented.
Key words: clustering, Kohonen map, metric, neural network, fuzzy sets, learning without a teacher.