Use of Posterior Probabilities to Evaluate Methods of Discriminant Analysis

By means of a case study the choice between several methods of discriminant analysis is presented. Experimental data of a two-groups problem with one or two variables is analysed. The different methods are compared according to posterior probabilities which can be computed for each subject and which are the basis of discriminant analysis. These posterior probabilities are analysed graphically as well as numerically.

Anhand einer Fallstudie wird beschrieben, wie unter mehreren Diskriminanzanalysemethoden ausgewählt werden kann. Dabei werden experimentelle Daten eines Zwei-Gruppen-Problems mit einem bzw. zwei Parametern analysiert. Mittels der A-posteriori-Wahrscheinlichkeiten werden verschiedene Methoden verglichen, die für jeden einzelnen Fall berechnet werden können und die den Ausgangspunkt jeder Dis-kriminanzanalyse bilden. Diese A-posteriori-Wahrscheinlichkeiten werden sowohl mit graphischen als auch mit numerischen Methoden analysiert.

Key-Words:

Linear Discriminant Analysis - Quadratic Discriminant Analysis - Discrimination with Kernel Methods - Posterior Probabilities - Error Rate - Penalty Scores

Schlüssel-Wörter:

Lineare Diskriminanzanalyse - quadratische Diskriminanzanalyse - Kernmethoden - A-posteriori-Wahrscheinlichkeiten - Fehlerraten - Strafpunkte

• References

• 1 Breiman L, Meisel W, Purcell E. Variable kernel estimates of multivariate densities. Technometrics 1977; 19: 135-144.
• 2 Buist A. S, Ross B. B. Predicted values for closing volume using a modified single-breath nitrogen test. Amer. Rev. resp. Dis 1973; 107: 744-752.
• 3 Habbema J. D. F, Hilden J, Bjerregaard B. The measurement of performance in probabilistic diagnosis. I. The problem, descriptive tools, and measures based on classification matrices. Meth. Inform. Med 1978; 17: 217-226.
• 4 Hermans J, Habbema J. D.F. Comparison of five methods to estimate posterior probabilities. EDV Med. Biol 1975; 6: 14-19.
• 5 Hermans J, Habbema J.D.F. Manual for the ALLOC discriminant analysis programs. Department of Medical Statistics; University of Leiden: 1976
• 6 Hilden J, Habbema J. D. F, Bjerregaard B. The measurement of performance in probabilistic diagnosis. II. Trustworthiness of the exact values of the diagnostic probabilities. Meth. Inform. Med 1978; 17: 227-237.
• 7 Hilden J, Habbema J. D. F, Bjerregaard B. The measurement of performance in probabilistic diagnosis. III. Methods based on continuous functions of the diagnostic probabilities. Meth. Inform. Med 1978; 17: 238-246.
• 8 Lende R. Van der, Hoygen C, Jansen-Koster E. J, Knijpstra S, Peset R, Visser B. F, Wolfs E. H. E, Orie N.G.M. A temporary decrease in the ventilatory function of an urban population during an acute increase in air pollution. Bull. Physiopath. resp 1975; 11: 31-43.
• 9 Lende R, van der Orie N.G.M. The MRC-ECCS questionnaire on respiratory symptoms. Use in epidemiology. Scand. J. resp. Dis 1972; 53: 218-226.
• 10 Nie N. H, Hull C. H, Jenkins J. G, Steinbremer K, Bent D. H. SPSS Statistical Package for the Social Sciences, second edition 1975, and SPSS Update. 1979. New York: Mcgraw-Hill.;
• 11 Raatgever J. W, Habbema J. D. F, Hermans J. Estimation of the kernel width in the variable kernel model. Technical Report. Departmet of Medical Statistics; University of Leiden: 1979
• 12 Sterk P. J, Quanjer P. H, Maas L. L. , Van der J, Wise M. E, van der Lende R. The validity of the single-breath nitrogen determination of residual volume. Bull, europ. Physiopath. resp 1980; 16: 195-213.
• 13 Victor N. Non-parametric allocation rules. In de Dombal F. T, Grémy F. (eds): Decision Making and Medical Care Amsterdam: Publ. Co. North-Holland; 1976: 515-527.