Hierarchical empowerment classification method and system for unbalanced data

Abstract

The invention provides a hierarchical empowerment classification method and system for unbalanced data. Comprising the following steps: an intra-class sub-concept cluster weight learning step: carrying out hierarchical clustering on training samples, and calculating the sample weight of each cluster according to the number of samples contained in each cluster in the clustering result, so that the sum of the number weights of each sub-concept cluster is kept consistent, and calculating the intra-class weight of each sub-concept cluster; and a global weight normalization step: normalizing the intra-class weight to a global weight according to a certain inter-class weight ratio, constructing a support vector machine with the weight as a classifier, and classifying the input non-equilibrium data with recognition by using the classifier. According to the method, the learning weight of the sub-concepts which are small in sample number and close to the classification surface can be improved, so that the recall rate of minority classes is improved, the false positive rate of the minority classes is reduced, and the overall comprehensive performance is ensured.

Description

technical field [0001] The invention belongs to the technical fields of machine learning, pattern recognition, etc., and in particular relates to a hierarchical weighting classification method and system for unbalanced data. Background technique [0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art. [0003] Class imbalance is very common in using machine learning to solve real-life problems. For example, in spam identification, the number of spam emails is much smaller than that of normal emails; in medical diagnosis, the number of patients is much smaller than the number of healthy people. Many traditional classification methods assume that the number of training data in each category is balanced, so that when the trained classifier is faced with unbalanced data, the classification results are more biased towards multi-sample categories. Although from the perspective of ov...

AI Technical Summary

Problems solved by technology

Although it can solve the unbalanced problem to a certain extent, the same processing method is used for all samples of a category, ignoring the distribution within the category, and blindly increasing the learning weight of the few-sample class, resulting in the false classification of the classifier after equalization for the few-sample class. Positive rate is higher

Moreover, it is also mentioned in the literature [Weiss G M, Provost F. Learning when training data are costly: The effect of class distribution on tree induction [J]. Journal of Artificial Intelligence Research, 2003, 19: 315-354.], due to "class Within-class imbalance phenomenon, classifiers usually have poor recognition effect on sub-concept clusters with less sample distribution

Images