Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Overtraining in neural networks that interpret clinical data

M L Astion1, M H Wener, R G Thomas

  • 1Department of Laboratory Medicine, University of Washington, Seattle 98195.

Clinical Chemistry
|September 1, 1993
PubMed
Summary

Overtraining in backpropagation neural networks reduces generalization ability. Monitoring cross-validation error, not training error, is crucial for optimal network performance in clinical data analysis.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Effects of individual and combined dietary weight loss and exercise interventions in postmenopausal women on adiponectin and leptin levels.

Journal of internal medicine·2013
Same author

Safety and efficacy of retreatment with a bioengineered hyaluronate for painful osteoarthritis of the knee: results of the open-label Extension Study of the FLEXX Trial.

Osteoarthritis and cartilage·2011
Same author

Safety issues in the development of treatments for osteoarthritis: recommendations of the Safety Considerations Working Group.

Osteoarthritis and cartilage·2011
Same author

Lymphoma and luekemia in rheumatoid arthritis: are they associated with azathioprine, cyclophosphamide, or methotrexate?

Journal of clinical rheumatology : practical reports on rheumatic & musculoskeletal diseases·2008
Same author

Prominent perivascular enhancement in primary central nervous system vasculitis.

Clinical and experimental rheumatology·2008
Same author

Primary central nervous system vasculitis: comparison of patients with and without cerebral amyloid angiopathy.

Rheumatology (Oxford, England)·2008

Area of Science:

  • Machine Learning
  • Artificial Intelligence
  • Clinical Informatics

Background:

  • Backpropagation neural networks (BNNs) are computational tools for pattern recognition in clinical data.
  • Unlike rule-based systems, BNNs learn from data through iterative training.
  • Overtraining is a key challenge where network generalization ability decreases with excessive training.

Purpose of the Study:

  • To analyze the phenomenon of overtraining in BNNs.
  • To evaluate the effectiveness of cross-validation in identifying optimal generalization points.
  • To assess BNN performance in differentiating giant cell arteritis (GCA) from other vasculitis (OTH).

Main Methods:

  • Utilized BNNs trained on clinical data from 807 patients (214 GCA, 593 OTH) with eight predictor variables.

Related Experiment Videos

  • Randomly assigned patient data into training (404 cases) and cross-validation (403 cases) sets.
  • Monitored training and cross-validation errors across eight independently trained networks.

    • Main Results:

    • Training error consistently decreased throughout the training process.
    • Cross-validation error typically minimized early in training, while training error continued to decline.
    • Training beyond the point of minimum cross-validation error led to increased cross-validation error, indicating reduced generalization.

    Conclusions:

    • Training error is an unreliable metric for assessing a BNN's generalization capability.
    • Cross-validation error monitoring is essential to determine the optimal training point for best network generalization.
    • This approach is vital for accurate clinical data interpretation using BNNs.