Input variable selection for multi-layer neural networks

Junichi Murata, Toru Nakazono, Kotaro Hirasawa

Research output: Contribution to journalArticle

Abstract

A method is proposed for selecting relevant input variables to multi-layer neural networks. A minimal set of inputs is selected which is necessary to obtain a network with a good generalization ability and some insight into the input-output relationship. The inputs of network are selected automatically by a combination of constructive and destructive algorithms. The constructive algorithm starts with a minimal input set and adds new inputs if necessary, while the destructive algorithm deletes unnecessary inputs. The main issue addressed here is the measure of input significance used in the constructive algorithm. Some measures are proposed based on mutual infomation and linear correlation paying much attention to the structural constraint imposed on the networks. The experimental results show that the measures are valid and that the derived network with the selected inputs has a good generalization ability.

Original languageEnglish
Pages (from-to)219-223
Number of pages5
JournalResearch Reports on Information Science and Electrical Engineering of Kyushu University
Volume3
Issue number2
Publication statusPublished - 1998

Fingerprint

Multilayer neural networks

All Science Journal Classification (ASJC) codes

  • Electrical and Electronic Engineering
  • Hardware and Architecture
  • Engineering (miscellaneous)

Cite this

Input variable selection for multi-layer neural networks. / Murata, Junichi; Nakazono, Toru; Hirasawa, Kotaro.

In: Research Reports on Information Science and Electrical Engineering of Kyushu University, Vol. 3, No. 2, 1998, p. 219-223.

Research output: Contribution to journalArticle

@article{60c595575f614000a2d5be6e4f6766eb,
title = "Input variable selection for multi-layer neural networks",
abstract = "A method is proposed for selecting relevant input variables to multi-layer neural networks. A minimal set of inputs is selected which is necessary to obtain a network with a good generalization ability and some insight into the input-output relationship. The inputs of network are selected automatically by a combination of constructive and destructive algorithms. The constructive algorithm starts with a minimal input set and adds new inputs if necessary, while the destructive algorithm deletes unnecessary inputs. The main issue addressed here is the measure of input significance used in the constructive algorithm. Some measures are proposed based on mutual infomation and linear correlation paying much attention to the structural constraint imposed on the networks. The experimental results show that the measures are valid and that the derived network with the selected inputs has a good generalization ability.",
author = "Junichi Murata and Toru Nakazono and Kotaro Hirasawa",
year = "1998",
language = "English",
volume = "3",
pages = "219--223",
journal = "Research Reports on Information Science and Electrical Engineering of Kyushu University",
issn = "1342-3819",
publisher = "Kyushu University, Faculty of Science",
number = "2",

}

TY - JOUR

T1 - Input variable selection for multi-layer neural networks

AU - Murata, Junichi

AU - Nakazono, Toru

AU - Hirasawa, Kotaro

PY - 1998

Y1 - 1998

N2 - A method is proposed for selecting relevant input variables to multi-layer neural networks. A minimal set of inputs is selected which is necessary to obtain a network with a good generalization ability and some insight into the input-output relationship. The inputs of network are selected automatically by a combination of constructive and destructive algorithms. The constructive algorithm starts with a minimal input set and adds new inputs if necessary, while the destructive algorithm deletes unnecessary inputs. The main issue addressed here is the measure of input significance used in the constructive algorithm. Some measures are proposed based on mutual infomation and linear correlation paying much attention to the structural constraint imposed on the networks. The experimental results show that the measures are valid and that the derived network with the selected inputs has a good generalization ability.

AB - A method is proposed for selecting relevant input variables to multi-layer neural networks. A minimal set of inputs is selected which is necessary to obtain a network with a good generalization ability and some insight into the input-output relationship. The inputs of network are selected automatically by a combination of constructive and destructive algorithms. The constructive algorithm starts with a minimal input set and adds new inputs if necessary, while the destructive algorithm deletes unnecessary inputs. The main issue addressed here is the measure of input significance used in the constructive algorithm. Some measures are proposed based on mutual infomation and linear correlation paying much attention to the structural constraint imposed on the networks. The experimental results show that the measures are valid and that the derived network with the selected inputs has a good generalization ability.

UR - http://www.scopus.com/inward/record.url?scp=0032155946&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0032155946&partnerID=8YFLogxK

M3 - Article

AN - SCOPUS:0032155946

VL - 3

SP - 219

EP - 223

JO - Research Reports on Information Science and Electrical Engineering of Kyushu University

JF - Research Reports on Information Science and Electrical Engineering of Kyushu University

SN - 1342-3819

IS - 2

ER -