| 摘 要: 医疗、金融和社交网络等许多领域的数据集通常存在大量的缺失值,这给数据分析和模型训练带来了巨大的挑战。文章提出一种基于高级生成对抗网络(Generative Adversarial Networks,GAN)的架构,用于稀疏数据集中的数据填充任务。该架构通过结合生成对抗网络的强大生成能力和深度学习技术,旨在更准确地填补数据集中的缺失值,从而提升数据质量和后续分析的准确性。实验结果表明,该框架的填补平均准确率约为85%,而相较于均值填补方法,其准确率提升约10.2百分点。在稀疏数据集上均取得显著的提升效果,为稀疏数据处理提供了新的解决方案。 |
| 关键词: 生成对抗网络(GAN);数据填充;稀疏数据集 |
|
中图分类号: TP399
文献标识码: A
|
|
| Application of Advanced Generative Adversarial Network Architecture for Data Imputation in Sparse Datasets |
|
WANG Shuo
|
(Dalian Neusoft University of Information, Dalian 116023, China)
wangshuo@neusoft.edu.cn
|
| Abstract: In many domains, such as healthcare, finance, and social networks, datasets often contain a large number of missing values, posing significant challenges for data analysis and model training. This paper proposes an advanced Generative Adversarial Network (GAN)-based architecture for the data imputation task in sparse datasets. By combining the powerful generative capabilities of GAN with deep learning techniques, the proposed architecture aims to impute missing values more accurately, thereby improving data quality and the accuracy of subsequent analyses. Experimental results demonstrate that the framework achieves an average imputation accuracy of approximately 85% which is about 10.2 percentage points higher than that of the mean imputation method. Significant improvements are observed across sparse datasets, providing a novel solution for sparse data processing. |
| Keywords: Generative Adversarial Networks (GAN); data imputation; sparse datasets |
