【摘要】：針對傳統(tǒng)過抽樣容易出現(xiàn)數(shù)據(jù)冗余和局限于處理靜態(tài)數(shù)據(jù)的問題,提出一種基于流形排序的動態(tài)過抽樣方法。該方法采用流形結(jié)構(gòu)描述數(shù)據(jù),根據(jù)數(shù)據(jù)內(nèi)在的全局流形結(jié)構(gòu)對少數(shù)類數(shù)據(jù)進(jìn)行排序,選擇出排序值高的數(shù)據(jù)執(zhí)行重采樣策略,以達(dá)到改善數(shù)據(jù)平衡度的目的。實(shí)驗(yàn)結(jié)果表明,在動態(tài)的不平衡數(shù)據(jù)集上,該方法獲得了比當(dāng)前同類方法更好的分類效果,還能有效提升分類器對少數(shù)類的識別性能。
[Abstract]:A dynamic oversampling method based on manifold sorting is proposed to solve the problem that data redundancy is easy to occur in traditional oversampling and it is limited to deal with static data. The method uses manifold structure to describe the data, sorts a few kinds of data according to the global manifold structure of the data, and selects the data with high sorting value to carry out resampling strategy, so as to improve the data balance. The experimental results show that the proposed method has better classification performance than the current similar methods on dynamic unbalanced datasets and can effectively improve the recognition performance of the classifier for a few classes.
【作者單位】： 東北電力大學(xué)信息工程學(xué)院;吉林供電公司信息通信分公司;
【分類號】：O212.2
，

本文編號：2414264