CONFIDENCE IN THE RESULTS OF LEARNING MACHINES TRAINED ON MASS-SPECTRA

被引：5

作者：

RICHARDS, JA ^{[1
]}

GRIFFITHS, AG ^{[1
]}

机构：

[1] DIGITAL EQUIPMENT AUSTRALIA PTY LTD, BRISBANE, Qld, AUSTRALIA

来源：

ANALYTICAL CHEMISTRY | 1979年 / 51卷 / 09期

关键词：

D O I：

10.1021/ac50045a006

中图分类号：

O65 [分析化学];

学科分类号：

070302 ; 081704 ;

摘要：

A measure is presented by which a user can assess the degree of confidence he can place on the results obtained by using a learning machine for the (multicategory) classification of mass spectra. Being based upon the weighted average of a set of backward probabilities, this measure is properly highly testing set dependent; its minimum value therefore is recommended as a reliable confidence indicator. Numerical results are presented to illustrate the significance of this indicator compared to measures often used to describe the integrity of results obtained with a learning machine. © 1979, American Chemical Society. All rights reserved.

引用

页码：1358 / 1361

页数：4

共 6 条

[1]

ABRAMSON N, 1963, INFORMATION THEORY C

[2]

GRIFFITHS A, 1977, THESIS J COOK U

[3] COMPUTERIZED LEARNING MACHINES APPLIED TO CHEMICAL PROBLEMS - MULTICATEGORY PATTERN CLASSIFICATION BY LEAST SQUARES [J].

KOWALSKI, BR ;

JURS, PC ;

ISENHOUR, TL ;

REILLEY, CN .

ANALYTICAL CHEMISTRY, 1969, 41 (06) :695-&

[4]

Nilsson N.J., 1965, LEARNING MACHINES

[5] CRITERIA FOR EVALUATION OF CLASSIFIERS FOR AUTOMATIC INTERPRETATION OF SPECTRA (PATTERN-RECOGNITION) [J].

ROTTER, H ;

VARMUZA, K .

ORGANIC MASS SPECTROMETRY, 1975, 10 (10) :874-884

[6] EVALUATION AND COMPARISON OF PATTERN CLASSIFIERS FOR CHEMICAL APPLICATIONS [J].

SOLTZBERG, LJ ;

WILKINS, CL ;

KABERLINE, SL ;

LAM, TF ;

BRUNNER, TR .

JOURNAL OF THE AMERICAN CHEMICAL SOCIETY, 1976, 98 (23) :7139-7144

← 1 →