Advances and Applications in Statistics
Volume 7, issue 2, Pages 255 - 280
(August 2007)
|
|
ASSESSMENT AND COMPARISON OF DISCRIMINANT ANALYSIS METHODS AND NON-PARAMETRIC ERROR ESTIMATORS WITH MIXED DATA
Ferran Galᮠ(Spain), Francesc Oliva (Spain) and Joan Guತia (Spain)
|
Abstract: In the field of social sciences and health, it is frequent that the information available comes from mixed data, i.e., a mixture of qualitative and quantitative variables. In this context, it is not easy to discriminate patterns: published studies show that no method is superior to the rest and it seems risky to base the decision solely by means of the error of the training data. In the present study we have identified interactive behaviors between the variables distribution and the classification error of several discriminant rules, as well as the bias and the mean square error from non-parametric error estimators. The protocol of the simulation study, dealing with mixed random vectors with a controlled dependence structure, approaches both problems at once and permits to extract relevant information. |
Keywords and phrases: discriminant analysis, non-parametric error estimators, mixed data. |
|
Number of Downloads: 361 | Number of Views: 984 |
|