Simulation of Behavior and Intelligence

On the value of features for classification

V. A. Kulyabko, I. S. Kuropteva, V. I. Serdobol'skii

Moscow

Abstract: Conditions are analyzed under which with the number of independent features high enough the likelihood relation logarithm can be normally approximated. An extreme experession is found for the mean value of the number of errors in classification in terms of features with different distributions. Varying that expression leads to asymptotic estimates of the contribution of features to reliability of the classification.

UDC: 025.4


Received: 25.02.1977

 English version: Automation and Remote Control, 1978, 39:3, 406–410

Bibliographic databases:

•