### Abstract

A computer system has been developed for the interactive design of pattern classifiers. The purpose of this system is to allow the researcher to interact with a data base of samples and incrementally design practical pattern recognizers, particularly for diagnostic applications. In this system, several well-known methods have been implemented, such as: Bayes' rule with independence and with 1st order dependence, and the K-nearest neighbor method. Adjustments can be made to the methods and the data base such that: only a subset of the features may be considered, decision boundaries varied, and misclassified patterns displayed. Frequencies and probability estimates for combinations of patterns of features are generated from the data base. In some cases, this leads to an effective alternative to the pattern recognition methods cited above. With a sufficiently large data base and a relatively small subset of features, complete dependence among features can sometimes be extracted. Probability estimates can be generated for combinations of patterns which cover all mutually exclusive possibilities. The design of a classifier is described which evaluates the prognosis of patients with bronchial asthma who are being considered for immunotherapy.

Original language | English (US) |
---|---|

Pages | 58-62 |

Number of pages | 5 |

State | Published - Jan 1 1978 |

Event | Proc Int Conf Cybern Soc Tokyo, Jpn, Nov 3-5 - Kyoto, Jpn Duration: Nov 7 1978 → Nov 7 1978 |

### Other

Other | Proc Int Conf Cybern Soc Tokyo, Jpn, Nov 3-5 |
---|---|

City | Kyoto, Jpn |

Period | 11/7/78 → 11/7/78 |

### Fingerprint

### All Science Journal Classification (ASJC) codes

- Engineering(all)

### Cite this

*INTERACTIVE SYSTEM FOR THE DESIGN OF CLASSIFIERS IN DIAGNOSTIC APPLICATIONS.*. 58-62. Paper presented at Proc Int Conf Cybern Soc Tokyo, Jpn, Nov 3-5, Kyoto, Jpn, .

}

**INTERACTIVE SYSTEM FOR THE DESIGN OF CLASSIFIERS IN DIAGNOSTIC APPLICATIONS.** / Weiss, Sholom M.; Kern, Kevin B.; Kulikowski, Casimir; Pincus, William.

Research output: Contribution to conference › Paper

TY - CONF

T1 - INTERACTIVE SYSTEM FOR THE DESIGN OF CLASSIFIERS IN DIAGNOSTIC APPLICATIONS.

AU - Weiss, Sholom M.

AU - Kern, Kevin B.

AU - Kulikowski, Casimir

AU - Pincus, William

PY - 1978/1/1

Y1 - 1978/1/1

N2 - A computer system has been developed for the interactive design of pattern classifiers. The purpose of this system is to allow the researcher to interact with a data base of samples and incrementally design practical pattern recognizers, particularly for diagnostic applications. In this system, several well-known methods have been implemented, such as: Bayes' rule with independence and with 1st order dependence, and the K-nearest neighbor method. Adjustments can be made to the methods and the data base such that: only a subset of the features may be considered, decision boundaries varied, and misclassified patterns displayed. Frequencies and probability estimates for combinations of patterns of features are generated from the data base. In some cases, this leads to an effective alternative to the pattern recognition methods cited above. With a sufficiently large data base and a relatively small subset of features, complete dependence among features can sometimes be extracted. Probability estimates can be generated for combinations of patterns which cover all mutually exclusive possibilities. The design of a classifier is described which evaluates the prognosis of patients with bronchial asthma who are being considered for immunotherapy.

AB - A computer system has been developed for the interactive design of pattern classifiers. The purpose of this system is to allow the researcher to interact with a data base of samples and incrementally design practical pattern recognizers, particularly for diagnostic applications. In this system, several well-known methods have been implemented, such as: Bayes' rule with independence and with 1st order dependence, and the K-nearest neighbor method. Adjustments can be made to the methods and the data base such that: only a subset of the features may be considered, decision boundaries varied, and misclassified patterns displayed. Frequencies and probability estimates for combinations of patterns of features are generated from the data base. In some cases, this leads to an effective alternative to the pattern recognition methods cited above. With a sufficiently large data base and a relatively small subset of features, complete dependence among features can sometimes be extracted. Probability estimates can be generated for combinations of patterns which cover all mutually exclusive possibilities. The design of a classifier is described which evaluates the prognosis of patients with bronchial asthma who are being considered for immunotherapy.

UR - http://www.scopus.com/inward/record.url?scp=0018059652&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0018059652&partnerID=8YFLogxK

M3 - Paper

AN - SCOPUS:0018059652

SP - 58

EP - 62

ER -