SUMMARY
The problem of training sample reduction is considered. A method for data reduction based on association rules is developed. The proposed method of training sample dimensionality reduction includes stages of reduction of instances, features and redundant terms, to evaluate the informativety of features uses the information about the extracted association rules. The developed method allows to create a partition of the feature space with less examples than in the original sample, which in turn allows the synthesis of simpler and more convenient for the perception of the diagnostic model. The practical value of these results is that on the basis of the proposed method the practical problem of reducing the training sample for the synthesis of the diagnostic model of quality confectionery products is solved.