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

A feed-forward network for input that is both categorical and quantitative.

Roelof K Brouwer1

  • 1Department of Computing Science, University College of the Cariboo, Kamloops, BC, Canada V2C 5N3. rkbrouwer@ieee.org

Neural Networks : the Official Journal of the International Neural Network Society
|December 16, 2003
PubMed
Summary

This study introduces a novel method for training multi-layer perceptrons (MLPs) with categorical data. The proposed approach effectively handles categorical variables, outperforming conventional methods in approximating continuous functions.

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

A discrete fully recurrent network of max product units for associative memory and classification.

International journal of neural systems·2002
See all related articles

Area of Science:

  • Machine Learning
  • Artificial Intelligence
  • Data Science

Background:

  • Multi-layer perceptrons (MLPs) typically require continuous input variables.
  • Categorical variables introduce discontinuities, posing challenges for standard MLP training.
  • Existing methods like numerical replacement or one-out-of-n encoding have limitations.

Purpose of the Study:

  • To develop a more effective method for training MLPs with mixed-type data (continuous and categorical).
  • To address the limitations of conventional approaches in handling categorical inputs.
  • To improve the accuracy and robustness of MLPs when dealing with discontinuous relationships introduced by categorical variables.

Main Methods:

  • Segregating categorical variables from continuous independent variables.

Related Experiment Videos

  • Training the MLP with multiple outputs, where each output unit corresponds to a unique combination of categorical values.
  • Utilizing a 'do not care' state for output units not associated with the current categorical input combination during training.

    • Main Results:

    • The proposed method significantly outperforms the conventional approach of assigning continuous values to categorical features.
    • For datasets with multiple categorical variables, the new method is more effective than one-out-of-n encoding.
    • Demonstrated superior performance across three comparative datasets.

    Conclusions:

    • The segregation approach offers a more effective way to incorporate categorical variables into MLPs.
    • This method provides a robust solution for handling discontinuous relationships in input data.
    • The findings suggest a new standard for preprocessing categorical data in neural network applications.