Criteria of a Selection of Attributes: Minimum of Mistakes Versus FRiS-function. Nikolay Zagoruiko, Irina Borisova, Olga Kutnenko. IWIM, Prague, 2007.
Article (in pdf)
Abstract. For an estimation of informativeness of separate attributes or their subsystems it is offered to use average value normalized functions of a rival similarity (FRiS) objects of training sample to the pattern. This criterion differs from criterion in the form of number of correctly recognized objects higher connection with results of recognition of control sample, a greater noise stability, an opportunity to estimate suitability of the chosen attributes and reliability of recognition of control object.
Keywords. Function of rival similarity, minimum of mistakes, feature selection, suitability of attributes.
References.
- Voronin Yu.A. The fundamentals the theory of similarity. Novosibirsk: Edition of Computer Centre, 1989.
- Zagoruiko N.G. Applied methods of the data and knowledge mining. Novosibirsk: Edition of Institute of Mathematics the Siberian Branch of the Russian Academy of Science, 1999.
- Kolmogorov A.N. To a question on suitability of the forecast formula founded by statistical way. - Factory laboratory. 1933. №1. pp. 164-167.
