Skip directly to search Skip directly to A to Z list Skip directly to page options Skip directly to site content

NIOSHTIC-2 Publications Search

Search Results

Analysis of injury risk factors with neural networks.

Authors
Landsittel-D; Arena-VC
Source
NOIRS 1997 Abstracts of the National Occupational Injury Research Symposium 1997. Washington, DC: National Institute for Occupational Safety and Health, 1997 Oct; :55
NIOSHTIC No.
20027742
Abstract
Neural networks have been shown to be a valuable analysis tool in research areas such as information science. Recent work has indicated that neural nets may also be useful in prediction and classification of biostatistical data. The importance of this research is to outline and illustrate how neural networks can be implemented to assess the statistical significance of injury risk factors and predict the probability of injury given individual characteristics. The output of the network is interpretable as the probability of an outcome, such as injury. The likelihood ratio test is used, in conjunction with a selection algorithm, to assess the statistical significance of injury risk factors. An overview of traditional methods for model selection and classification of injury data is presented and the advantages and disadvantages of using neural networks to analyze injury risk factors are examined. Neural networks are a type of nonlinear regression where the nature of the association between the covariates and outcome is not explicitly specified. The form of, and interaction between variables are also implicitly fit in the model. Both positive and negative implications of the previous statements are investigated. In cases where the nature of the association between injury and possible causal factors is unknown, neural networks may provide an effective alternative to standard methods. Since neural nets depend on an iterative routine to solve for the optimal weights, the solution may only represent a local minimum and the fact that a particular association is not explicitly specified may lead to over fitting the data. These and other possible limitations of neural nets are explored specifically with respect to analysis of injury risk factors. Several methods to improve the performance of neural networks, including weight decay, committees of networks, and cross validation are illustrated. General conclusions about the role of neural networks in injury risk factor analysis are made.
Keywords
Injuries; Risk-analysis; Risk-factors; Biostatistics; Models; Mathematical-models
Publication Date
19971015
Document Type
Conference/Symposia Proceedings; Abstract
Fiscal Year
1998
NTIS Accession No.
NTIS Price
NIOSH Division
DSR
Source Name
NOIRS 1997 Abstracts of the National Occupational Injury Research Symposium
State
WV
TOP