Support vector data description method for fuzzy zone negative class samples based on class center distance

Abstract

The invention provides a support vector data description method for fuzzy zone negative class samples based on a class center distance, and the method comprises the steps: firstly providing a data set for training testing, and obtaining the central point of this class through calculating the mean value of all samples of this class; secondly calculating a second normal form distance between the samples and the other class center, calculating a second normal form distance between the samples and the center of this class, solving a proportion of the two distances, and obtaining the classification capability values of the samples; thirdly obtaining a corresponding membership degree through employing the fuzzy classification capability value of each sample, adding the membership degree items to two types of support vector data description target functions, reconstructing the target functions, and obtaining an algorithm target function; fourthly solving the target function through employing a Lagrange dual form, obtaining a secondary planning form function, and solving the circle center of a classification boundary; finally classifying testing samples, and obtaining classification precision. Compared with a conventional support vector data description data, the method provided by the invention can obtain the classification capability values of the samples through employing the proportion of the second normal form distance between the samples and the other class center to the second normal form distance between the samples and the class center, adds the membership degree so as to distinguish contribution degrees of different samples, highlights the importance of the boundary samples, and improves the classification accuracy.

Description

technical field [0001] The invention belongs to the field of pattern recognition, in particular to a method for describing support vector data of fuzzy samples with negative samples based on centroid distance. Background technique [0002] The training samples of traditional pattern classification methods generally require samples of multiple categories, so two-class and multi-class classifiers are designed. In some special fields, such as machine fault diagnosis, disease analysis, intrusion detection, friend-or-foe identification, and credit card fraud, we will encounter only one class of samples that can be used to train classifiers, and no other classes of samples participate. Because in these fields, it is difficult for us to obtain multi-class samples or the cost of obtaining samples of other classes is relatively high. The boundary of the two-class classifier requires two types of instance samples to support, which will cause us to be unable to use two types of instan...

AI Technical Summary

Problems solved by technology

[0004] In order to solve the existing technical problems, the present invention not only considers that different types of data samples have different location information, but also considers the degree of influence of the same type of data samples on the decision boundary, and emphasizes the contribution of different samples. Support vector data description method for fuzzy samples with negative classes based on centroid distance to effectively deal with two types of problems

Images