Utility and Risk Evaluation of Synthetic Data by Orthogonal Transformation
Utility and Risk Evaluation of Synthetic Data by Orthogonal Transformation
复制标题
正交变换合成数据的效用和风险评估
DOI：
10.1007/s12626-022-00107-x
复制
发表时间：
2022
期刊：
影响因子：
--
通讯作者：
Sano Natsuki
中科院分区：
文献类型：
--
作者：
Jun-ichi Itaya;Atsue Mizushima;and Kengo Kurosaka;Daiki Maki;Daiki Maki;Daiki Maki;Takahiro ITO;Sano Natsuki
在统计披露控制中发布合成数据使得识别个人记录变得困难，因为合成数据与原始数据不同。我们提出了一种使用正交变换生成合成数据的方法，以及由此生成的数据的效用度量。该方法可以根据所提出的效用度量来控制生成的合成数据的效用。我们将该方法应用于从日本家庭收入和支出全国调查中获得的匿名数据，并为其生成综合数据。此外，我们评估了由此生成的数据的效用和风险，并将其与通过其他方法生成的合成数据进行了比较。我们发现，通过调整采用的特征值的数量，所提出的方法可以生成比其他方法具有更高效用的合成数据。
Releasing synthetic data in statistical disclosure control makes identifying individual records difficult, as synthetic data differ from the original data. We propose a method for generating synthetic data using orthogonal transformation, along with a utility measure for the data thus generated. This method can control the utility of the generated synthetic data in terms of the proposed utility measure. We applied the method to anonymized data, obtained from a national survey of family income and expenditure in Japan, and generated synthetic data for it. Additionally, we evaluated the utility and risk of the data thus generated and compared them with synthetic data generated through other methods. We find that the proposed method can generate synthetic data with a higher utility than other methods by adjustment of the number of adopted eigen value.