使用NGN算法改进不平衡数值数据的研究

打开文本图片集

Research on improving imbalanced numerical data using NGN algorithm

Xing Changzheng，Zheng Xin，LiangJunfeng （CollegeofElectronic& Information Engineering，Liaoning Technical University，HuludaoLioning1251o5，China）

Abstract：When minorityclassamplesare scarce，traditional oversampling methods struggleto increasethesamplecount. This paper introduced a NGN algorithm that synthesized new data byadding generator-generated dataas noise to theoriginal minorityclassamplesuntilbalance wasachieved.Thegenerator employedafour-layerfullyconnectednetworkandintegrated low-structureandhigh-structurefeaturegenerationtechniquestoenhancethequalityanddiversityofthegenerateddata.For verylimited minorityclassamples，NGNgeneratednewsamples，mergedthemwiththeoriginal minorityclassdata，and performedclustering to achieve balance withinclusters while minimizing the impactof noise.The study evaluated NGNon6unbalanceddatasets，applied4oversamplingalgorithms tobalancethedatasets，andclasifiedthebalanceddatasetsusing4classificationmethods.TheexperimentalresultsdemonstratethatNGNefectivelyincreasesthenumberof minorityclasssamples， enhances the model’sability to learn minority classfeatures，and significantly improves classification performance.

Key words：numerical generator network（NGN）；generator；noise；extremely scarce minority class；balance

0 引言

数据不平衡的问题源自于样本分布的不均衡。（剩余15909字）