Binary Classification of Objects with Nominal Indicators

In this work a problem is studied of classification of respondents into classes accepting and not participation in a charity actions. An optimal (in Bayes sense) decisive discriminant rule of division of objects on two classes is constructed for the case when all indicators of observable objects are measured in a nominal scale, and there are signs of dependence between them . Using ROC-analysis methods, comparison of the developed rule with a rule implemented in the software package SPSS (Fisher’s discriminant rule), «naive» Bayesian classifier, a rule based on support vector machines (SVM) method and implemented in SPSS package binary logistic regression classifier is made. Results of the ROC-analysis have shown that the proposed rule has higher quality than all other mentioned rules of classification of respondents.