A comparison of bayes discrimination, location model and logistic regression on the basis of real

The paper compares alternative methodologic approaches for two-group discrimination with mixed explanatory variables. The important problem is to select the best discriminant method for the model. For this purpose on the basis of the data connected with recognizing childhood asthma we compare functioning of different methods. We have applied various discriminant methods based on mixed quantitative and qualitative observations. The discrimination is performed on the grounds of all mixed variables and also on the basis of the most discriminating variables. The classical methods and nonparametric ones, such as kernel (with different kernel functions) and the nearest neighbour (with different distances), are employed. We have also applied the logistic discrimination and the location linear model of the first order. For the explored examples of mixed model - nonparametric kernel methods wit h different covariance matrices gave the best results.