Record Display for the EPA National Library Catalog


Main Title Monte Carlo Studies of the Classifications Made by Nonparametric Linear Discriminant Functions.
Author Stouch, T. R. ; Jurs, P. C. ;
CORP Author Pennsylvania State Univ., University Park.;Health Effects Research Lab., Research Triangle Park, NC.
Year Published 1985
Report Number EPA/600/J-85/320;
Stock Number PB86-163847
Additional Subjects Pattern recognition ; Classifications ; Monte Carlo method ; Reprints ; Discriminant functions
Library Call Number Additional Info Location Last
NTIS  PB86-163847 Some EPA libraries have a fiche copy filed under the call number shown. 07/26/2022
Collation 8p
Chance factors in pattern recognition studies utilizing nonparametric linear discriminant functions are examined. The relationship between complete linear separation of a data set and the dimensionality of the study is well-known. Also, due to the nature of the inequalities from which these numerical techniques are derived, 50% separation is always assured. The paper investigates the probability of achieving less than 100% but greater than 50% chance separations as a function of the dimensionality and class membership distribution. It is shown that the fraction of correct classifications due to chance factors increases dramatically as the dimensionality of the study increases. These results serve to redefine the level of expected chance classifications as a function of the number of observations, the dimensionality, and the class membership distributions. The results can be used to assess the classification results obtained with a given linear discriminant function.