Optimal division of dataset into three subsets for Artificial Neural Network models

Goloka Behari Sahoo, Chittaranjan Ray

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

The generalization ability of Artificial Neural Networks (ANNs) is significantly undermined if the datasets presented to ANN for training do not contain sufficient information in all dimensions of the modeling domain. Under practical circumstances, it is not possible to get a large dataset for ANN use. This paper presents a systematic approach that answers the questions "which samples" and "how many samples" should be selected for the datasets required by ANN when available data are limited. The predictive ability of ANN largely depends on the network's structure and internal parameter. Although some guidance is available in the literature for the choice of geometry and internal parameters, the solution space lies in large range and thus their combination is so large. This paper presents the use of micro genetic algorithms (μGA) to develop a μGA-ANN model to search for the optimal combination of ANN geometry and internal parameters.

Original languageEnglish (US)
Title of host publicationProceedings of the 3rd Indian International Conference on Artificial Intelligence, IICAI 2007
Pages859-872
Number of pages14
StatePublished - 2007
Externally publishedYes
Event3rd Indian International Conference on Artificial Intelligence, IICAI 2007 - Pune, India
Duration: Dec 17 2007Dec 19 2007

Publication series

NameProceedings of the 3rd Indian International Conference on Artificial Intelligence, IICAI 2007

Conference

Conference3rd Indian International Conference on Artificial Intelligence, IICAI 2007
Country/TerritoryIndia
CityPune
Period12/17/0712/19/07

Keywords

  • Back propagation neural network
  • Data division
  • Micro genetic algorithms
  • Optimization network geometry
  • Radial basis function network
  • Self-organizing map

ASJC Scopus subject areas

  • Artificial Intelligence

Fingerprint

Dive into the research topics of 'Optimal division of dataset into three subsets for Artificial Neural Network models'. Together they form a unique fingerprint.

Cite this