Approbation of the data classification method based on the SVM algorithm and the k nearest neighbors algorithm

Abstract

The aim of the work is to test the classification method based on the combined use of SVM and kNN classifiers. At the first stage of the classification method, on the basis of the initial training dataset U, SVM classifier is developed and the width of the Ω area is determined. This Ω area contains all objects incorrectly classified by SVM classifier, which form together with correctly classified objects that have fallen into the Ω area and the corresponding object class labels from the Ω area a new dataset G. At the second stage of the classification method, the kNN classifier developed on the basis of information about the objects of the set U\\G is applied to all objects of the dataset G from the Ω area. The values of the parameters of the kNN classifier are determined experimentally in such a way as to ensure the highest possible accuracy of object classification. Since the correctly classified objects can also enter the Ω area formed in the above way, the condition for the applicability of the proposed method is a general improvement in the classification quality. The presented results of experimental studies confirm the effectiveness of the proposed method in the problem of classifying complex multidimensional data.