A. V. Gavrilov, V. M. Kangler, S. A. Zaitsev


Novosibirsk State Technical University,
Novosibirsk, Russia, e-mail:


The report considers the program for database analysis using neural networks. This program is new version of program described in [1]. The program is intended for solving of two tasks:

- Researching of various of neural networks models. Mainly, parameters influence on networks training and working study.

- Using neural networks for performing analysis.

Within area of the first task, the program allows working with a multitude neural networks. The neural networks are added to main program as DLL-implemented plug-ins. In order to interact with neural networks a shell uses a universal interface that allows working virtually with any neural network model. Thus, the program is expandable and allows learning news neural networks models.

The program allows performing four types of analysis:

    1. Prediction
    2. Recognition
    3. Clusterization
    4. Associative search

In order to increase neural networks work efficiency, the shell allows data preprocessing. At resolving prediction tasks a trend can be used, that is an analogue of the first derivate and this allows to predict non-periodical data.

It is possible to use encoding of segments in order to reach more efficient processing of fractional numbers; this makes impact from stochastic inaccuracies in input data lower.

This investigation in part is supported by grant of Ministry of Education of Russia Federation.


[1] A.V.Gavrilov, V.M.Kangler. The use of Artificial Neural Networks for Data Analysis // The Third Russian-Korean International Symposium on Science and Technology. - Novosibirsk: NSTU, 1999. - Proceedings/ - Vol.1. - P.257-260; Abstracts. - Vol. 1. - P. 192.