A train set forming for using artificial neural networks to database errors search

Abstract

A train set forming for using artificial neural networks to database errors search

V.V. Galushka, V.A. Fatkhi

Incoming article date: 03.05.2013

It describes a method of train set forming for using artificial neural networks to search inauthentic rows  in databases tables. An existing methods of the reliability ensure involve the use of integrity constraints and provides a truthfulness, but there is still a possibility of entering of inauthentic data, appropriate to all constraints. A more accurate assessment of reliability is possible with the use of artificial neural networks that require a training set. The main requirement for the training set - representation includes sufficiency, diversity and evenness. The approaches to each of these requirements are describes. Also calculations a sufficient number of rows for training neural networks of various types is given, as well as the results of experiments that confirm correctness of the theoretical calculations.

Keywords: database, authenticity, artificial neural networks, training set, representation