Using genetic algorithms to select and create features for pattern classification. Technical report
Abstract
Genetic algorithms were used to select and create features and to select reference exemplar patterns for machine vision and speech pattern classification tasks. On a 15-feature machine-vision inspection task, it was found that genetic algorithms performed no better than conventional approaches to feature selection but required much more computation. For a speech recognition task, genetic algorithms required no more computation time than traditional approaches but reduced the number of features required by a factor of five (from 153 to 33 features). On a difficult artificial machine-vision task, genetic algorithms were able to create new features (polynomial functions of the original features) that reduced classification error rates from 10 to almost 0 percent. Neural net and nearest-neighbor classifiers were unable to provide such low error rates using only the original features. Genetic algorithms were also used to reduce the number of reference exemplar patterns and to select the value of k for a k-nearest-neighbor classifier. On a .338 training pattern vowel recognition problem with 10 classes, genetic algorithms simultaneously reduced the number of stored exemplars from 338 to 63 and selected k without significantly decreasing classification accuracy. In all applications, genetic algorithms were easy to apply and found good solutions inmore »
- Authors:
- Publication Date:
- Research Org.:
- Massachusetts Inst. of Tech., Lexington, MA (United States). Lincoln Lab.
- OSTI Identifier:
- 5439322
- Report Number(s):
- AD-A-235165/8/XAB; TR-892
CNN: F19628-90-C-0002
- Resource Type:
- Technical Report
- Country of Publication:
- United States
- Language:
- English
- Subject:
- 99 GENERAL AND MISCELLANEOUS//MATHEMATICS, COMPUTING, AND INFORMATION SCIENCE; PATTERN RECOGNITION; ALGORITHMS; ACCURACY; CLASSIFICATION; COMPUTERS; ERRORS; FUNCTIONS; NEURAL NETWORKS; PARALLEL PROCESSING; POLYNOMIALS; REDUCTION; SPEECH; VISION; CHEMICAL REACTIONS; MATHEMATICAL LOGIC; PROGRAMMING; 990200* - Mathematics & Computers
Citation Formats
Chang, E I, and Lippmann, R P. Using genetic algorithms to select and create features for pattern classification. Technical report. United States: N. p., 1991.
Web.
Chang, E I, & Lippmann, R P. Using genetic algorithms to select and create features for pattern classification. Technical report. United States.
Chang, E I, and Lippmann, R P. 1991.
"Using genetic algorithms to select and create features for pattern classification. Technical report". United States.
@article{osti_5439322,
title = {Using genetic algorithms to select and create features for pattern classification. Technical report},
author = {Chang, E I and Lippmann, R P},
abstractNote = {Genetic algorithms were used to select and create features and to select reference exemplar patterns for machine vision and speech pattern classification tasks. On a 15-feature machine-vision inspection task, it was found that genetic algorithms performed no better than conventional approaches to feature selection but required much more computation. For a speech recognition task, genetic algorithms required no more computation time than traditional approaches but reduced the number of features required by a factor of five (from 153 to 33 features). On a difficult artificial machine-vision task, genetic algorithms were able to create new features (polynomial functions of the original features) that reduced classification error rates from 10 to almost 0 percent. Neural net and nearest-neighbor classifiers were unable to provide such low error rates using only the original features. Genetic algorithms were also used to reduce the number of reference exemplar patterns and to select the value of k for a k-nearest-neighbor classifier. On a .338 training pattern vowel recognition problem with 10 classes, genetic algorithms simultaneously reduced the number of stored exemplars from 338 to 63 and selected k without significantly decreasing classification accuracy. In all applications, genetic algorithms were easy to apply and found good solutions in many fewer trials than would be required by an exhaustive search. Run times were long but not unreasonable. These results suggest that genetic algorithms may soon be practical for pattern classification problems as faster serial and parallel computers are developed.},
doi = {},
url = {https://www.osti.gov/biblio/5439322},
journal = {},
number = ,
volume = ,
place = {United States},
year = {Mon Mar 11 00:00:00 EST 1991},
month = {Mon Mar 11 00:00:00 EST 1991}
}