UDC 004.855.5
TWO-STAGE DATA CLASSIFICATION METHOD BASED ON SVM-ALGORITHM AND THE k NEAREST NEIGHBORS ALGORITHM
L. A. Demidova, PhD (technical sciences), full professor, RSREU, Ryazan; This email address is being protected from spambots. You need JavaScript enabled to view it.
Yu. S. Sokolova, senior teacher, RSREU, Ryazan; This email address is being protected from spambots. You need JavaScript enabled to view it.
The classification problem of elaborate multidimensional data which is inherent in various socioeconomic, technical and other systems has been considered. The aim is the classification accuracy increase of elaborate multidimensional data by means of development of two-stage classification method based on the combined use of SVM and kNN classifiers. At the first stage of the classification method SVM classifier on the base of initial learning dataset U is developed and the width of Ω-area containing all objects classified erroneously by SVM classifier. Objects classified erroneously together with correctly classified objects which are also located in the Ω-area and the corresponding classes tags of objects from Ω-area form the new G dataset. At the second stage of the classification method kNN classifier developed on the base of information about the objects of U\G set is applied to all objects of G data set from Ω-area. In case of improvement of the classification quality of objects belonging to Ω-area, the offered two-stage method can be recommended for classification of new objects. The parameters values of kNN classifier are defined experimentally to provide the greatest possible classification accuracy of objects. As the correctly classified objects can also get to Ω-area created in the above-stated way, the condition of applicability of the offered method is general improvement of classification quality. The given results of experimental studies confirm the efficiency of the offered method application in the classification problem of elaborate multidimensional data.
Key words: SVM classifier, support vectors, kernel function type, kernel function parameters, regularization parameter, kNN classifier, classification method.