Instance-based method to extract rules from neural networks

Dae Eun Kim, Jaeho Lee

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review


It has been shown that a neural network is better than induction tree applications in modeling complex relations of input attributes in sample data. Those relations as a set of linear classifiers can be obtained from neural network modeling based on back-propagation. A linear classifier is derived from a linear combination of input attributes and neuron weights in the first hidden layer of neural networks. Training data are projected onto the set of linear classifier hyperplanes and then information gain measure is applied to the data. We propose that this can reduce computational complexity to extract rules from neural networks. As a result, concise rules can be extracted from neural networks to support data with input variable relations over continuous-valued attributes.

Original languageEnglish
Title of host publicationArtificial Neural Networks - ICANN 2001 - International Conference, Proceedings
EditorsKurt Hornik, Georg Dorffner, Horst Bischof
PublisherSpringer Verlag
Number of pages6
ISBN (Print)3540424865, 9783540446682
StatePublished - 2001
EventInternational Conference on Artificial Neural Networks, ICANN 2001 - Vienna, Austria
Duration: 21 Aug 200125 Aug 2001

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349


ConferenceInternational Conference on Artificial Neural Networks, ICANN 2001


Dive into the research topics of 'Instance-based method to extract rules from neural networks'. Together they form a unique fingerprint.

Cite this