Unbalanced data classification method based on clustering downsampling

Abstract

The invention discloses an unbalanced data classification method based on clustering downsampling. The unbalanced data classification method comprises the following steps: utilizing a quick search anddiscovery density peak value clustering algorithm for clustering various categories of samples of a training set to obtain a clustering result; dividing multiple categories of samples in the trainingset into N clusters; forming a new sample set by each cluster of sample of multiple categories of samples in the training set and small categories of samples in the training set, and utilizing a support vector machine for classification to obtain the support vectors of multiple categories of samples in the training set; extracting the new training set formed by the support vector of each clusterand the small categories of samples in the training set; training the new training set through the support vector machine, and carrying out performance evaluation through a cross verification set. Thetraining time of the classifier can be shortened, in addition, the identification rate of small categories of samples can be improved under a situation that the identification rate of multiple categories of identification samples is not harmed, and the performance of the classifier is improved.

Description

technical field [0001] The invention relates to the research field of pattern recognition, in particular to a classification method for unbalanced data based on clustering and downsampling. Background technique [0002] The classification problem is a very important research content in the fields of pattern recognition and machine learning. It has a very wide range of applications in real life, such as handwritten digit recognition in banking systems, face recognition in security monitoring systems, and network security. intrusion detection etc. At present, there are some relatively mature classification methods for dealing with classification problems, such as: decision tree, K-nearest neighbor, neural network, support vector machine and other methods. extensive attention. These traditional classification methods are all proposed based on the assumption of balanced class distribution, and their main purpose is to improve the overall classification performance, and show go...

AI Technical Summary

Problems solved by technology

However, the downsampling technology may lose representative multi-class sample information while deleting multi-class samples, and the classification surface will be shifted.

Images