Using feedforward neural networks and forward selection of input variables for an ergonomics data classification problem

  • Authors:

  • Chuen-Lung Chen;David B. Kaber;Patrick G. Dempsey

  • Affiliations:

  • Department of Management Information Systems, National Chengchi University, Taipei, Taiwan, 11623, E-mail: chencl@mis.nccu.edu.tw;Department of Industrial Engineering, North Carolina State University, Raleigh, NC 27695–7906, U.S.A., E-mail: dbkaber@eos.ncsu.edu;Liberty Mutual Research Center for Safety & Health, Hopkinton, MA 071748, U.S.A. E-mail: Patrick.Dempsey@@LibertyMutual.com

  • Venue:
  • Human Factors in Ergonomics & Manufacturing
  • Year:
  • 2004

Quantified Score

Hi-index 0.00

Visualization

Abstract

A method was developed to accurately predict the risk of injuries in industrial jobs based on datasets not meeting the assumptions of parametric statistical tools, or being incomplete. Previous research used a backward-elimination process for feedforward neural network (FNN) input variable selection. Simulated annealing (SA) was used as a local search method in conjunction with a conjugate-gradient algorithm to develop an FNN. This article presents an incremental step in the use of FNNs for ergonomics analyses, specifically the use of forward selection of input variables. Advantages to this approach include enhancing the effectiveness of the use of neural networks when observations are missing from ergonomics datasets, and preventing overspecification or overfitting of an FNN to training data. Classification performance across two methods involving the use of SA combined with either forward selection or backward elimination of input variables was comparable for complete datasets, and the forward-selection approach produced results superior to previously used methods of FNN development, including the error back-propagation algorithm, when dealing with incomplete data. © 2004 Wiley Periodicals, Inc. Hum Factors Man 14: 31–49, 2004.