基于WGAN的小样本建模方法研究

Research on Small Sample Modeling Method Based on WGAN

针对小样本建模存在的模型拟合效果欠佳、参数估计不准确的问题,利用生成对抗网络可以捕获原始数据分布且能够生成服从其分布的数据的特性,文章将生成对抗网络用于扩展小样本数据的规模,并对生成的数据进行优化处理,使用优化后的数据集进行多元回归分析。结果表明,模型拟合结果与原始数据相比效果更好。生成对抗网络可以作为扩大样本量的一种方法,应用于经济社会统计中。

In view of the fact that the model fitting effect of small sample modeling is not good, that parameter estimation is not accurate, and that the use of generative adversarial networks can capture the original data distribution and can generate data obeying its distribution, this paper uses generative adversarial networks to expand the scale of small samples, and optimizes the generated samples. Then the optimized data set is used for multiple regression analysis. The results show that the model fitting results are better than the original data. The generation of adduction network can be used in economic and social statistics as a method to enlarge the sample size.