摘要
针对时间序列数据中不平衡问题,提出一种基于样本边界和局部密度的生成对抗网络(density-boundary generative adversarial network,DBGAN)。根据样本分布特点,利用共享最近邻、局部离群因子以及高斯核函数获得少数类样本的重要性得分,指导生成器合成具有代表性的少数类样本。同时借鉴CatGAN模型对cGAN的判别器进行修改,使其能够适应多样性的类内分布。实验结果表明,该模型在不同数据集上的性能指标得到提升,有效缓解类不平衡问题。
Aiming at the imbalance problem in time series data,a density-boundary generative adversarial network(DBGAN)based on sample boundary and local density is proposed.Based on the distribution characteristics of samples,the proposed algorithm calculates importance scores for minority class samples using shared nearest neighbor,local outlier factor,and Gaussian kernel function,guiding the generator to synthesize representative minority class samples.Additionally,inspired by CatGAN,the discriminator of the conditional generative adversarial network(cGAN)is modified to adapt to diverse intra-class distributions.The experimental results show that the performance index of the model is improved on different datasets,effectively alleviating the class imbalance problem.
作者
杨瑞
张海清
李代伟
陈金京
任李娟
刘佳璇
YANG Rui;ZHANG Hai-qing;LI Dai-wei;CHEN Jin-jing;REN Li-juan;LIU Jia-xuan(College of Software Engineering,Chengdu University of Information Technology,Chengdu 610225,China;Data Ecology and Smart Space Lab,Sichuan Provincial Informatization Application Support Software Engineering and Technology Research Center,Chengdu 610255,China)
出处
《计算机工程与设计》
北大核心
2025年第8期2178-2185,共8页
Computer Engineering and Design
基金
国家自然科学基金项目(61602064)
四川省科技厅基金项目(2021YFH0107、2022YFS0544、2022NSFSC0571)
成都信息工程大学科技创新能力提升计划基金项目(KYQN202223)。
关键词
时间序列
类不平衡
生成对抗网络
共享最近邻
局部离群因子
边界样本
局部密度
time series
class imbalance
generative adversarial network
shared nearest neighbors
local outlier factor
boundary sample
local density
