Record Details

????? ???????? ???????????? ?? ??????? ??????? ?? ????????????

Електронний архів Житомирського державного технологічного університету

View Archive Info
 
 
Field Value
 
Title ????? ???????? ???????????? ?? ??????? ??????? ?? ????????????
Method of neural network model w?th real output construct?onon precedents
 
Creator ????????, ?.?.
Subbot?n, S.A.
 
Subject ???????? ??????
??????
???????
??????? ????????
??????????????
neural networks
feature selection
machine learning
diagnosis
 
Description ???????? ???????? ????????????? ??????? ??????? ?????????? ??????????? ?? ????????????.
????????????? ????? ???????? ??????? ?i???i???? ??????????? ?? ???????????? ?? ??????
????????? ?????, ?? ?????????? ???????i??i ?? ??????????? ??????????? ?????. ???????????
?? ????????? ????????????? ?????? ???????? ?????? ??? ?????? ???????????: ?????? ????
???????? ?????????? ?????????????? ?????????? ?? ???????? ? ???????? ?????, ??????? ????????
?????????? ?? ?????????? ???????????? ???????? ??? ?????????, ?????? ???????? ????? ?????????
?????? ?????????? ?? ?????????? ?????????????? ?????????? ?? ??????? ????????. ????????????
????????????? ?????? ???????? ???????? ???????????????? ??????????? ? ???????????
?????????? ?? ???????? ?????????????? ???????????. ????????? ????????????????
??????????? ???????????? ?????? ??? ????????? ?????????? ???????.
The problem of automation of quantitative dependence model synthesis on precedents is solved. The method of quantitative dependence model construction on pecedents based on neural networks taking into account the topological and statistical properties of the data is proposed. The neural network synthesized using the proposed method has block architecture: the first block defines belonging of recognized instance to the cluster in the feature space, a second block realize calculations for partial regression models for clusters, the third block selects partial model according to recognized instance belonging to a particular cluster. The using of the proposed method provides to receive the interpretable neural network models with minimal complexity and high generalization abilities. The experimental study of the developed method in solving practical problems is conducted.
 
Date 2016-03-31T10:47:00Z
2016-03-31T10:47:00Z
2013
 
Type Article
 
Identifier http://eztuir.ztu.edu.ua/123456789/2451
 
Language uk
 
Relation ?????? ????. ?????: ???????? ?????;4(67)
 
Publisher ????