High-dimensional imbalanced data classification method based on SVM

Abstract

The invention proposes a high-dimensional imbalanced data classification method based on SVM. The method includes two parts. The first part is feature selection. An SVM-BRFE algorithm is used to carryout boundary resampling to find the optimal feature weight to carry out feature importance measuring, feature selecting and training set updating, and the process is repeated. Finally, a feature mostconductive to enhancing the F1 value is retained, and other features are removed. A subsequent training process is carried out under the condition with feature redundancy and irrelevant feature combination as less as possible and dimension as low as possible. The influence of a high-dimensional problem on an imbalance problem and the constraint of an SMOTE oversampling algorithm are reduced. Thesecond part is data sampling. An improved SMOTE algorithm, namely PBKS algorithm, is used. Few classes in boundaries automatically partitioned by SVM are used as distance constraints in the Hilbert space Dxij<H>, and original constraints are replaced. A grid method is used to find the approximate preimage. The method provided by the invention can finish the classification task of high-dimensionalunbalanced data stably and effectively, and can obtain a considerable effect.

Description

technical field [0001] The invention belongs to the technical field of data classification, and in particular relates to a method for classifying unbalanced samples. Background technique [0002] In the classification task of data mining, the current classification methods for high-dimensional imbalanced data are to solve the high-dimensional problem or imbalance problem first, and then solve another problem, without considering the impact of high-dimensional characteristics on the classification of unbalanced data. Implications of new problems and imbalanced features for classification of high-dimensional data. The classification task of imbalanced data is mainly carried out from two levels: sampling at the data level and classification at the algorithm level. [0003] The sampling method at the data level is one of the important means to solve the unbalanced data distribution in the sample space. Through methods such as undersampling, resampling, and mixed sampling, the s...

AI Technical Summary

Problems solved by technology

However, in high-dimensional unbalanced data, due to the existence of high-dimensional problems, the traditional sampling method cannot change the classifier's emphasis on the majority class, thus making the traditional sampling method meaningless

The experimental research in [21] shows that although the SMOTE method can increase the classifier’s attention to minority classes in low-dimensional data, the effect is not obvious in high-dimensional data.

The main reason is that the minority class generated by the SMOTE method will introduce the correlation between samples in the new sample space, rather than the correlation between features, so the generated minority class cannot restore the original sample space well. class distribution

Images