基于单元间距离估计的不等概率抽样算法及应用

Unequal Sampling Algorithm Based on Kriging with Partial Least Squares Algorithm and Its Application

现有大数据的不等概率抽样方法大多基于数据分布,泛化能力较差。为此,利用多层感知机、XGBoost和Kriging模型估计总体单元间的相对距离,提出针对海量数据的不等概率抽样算法。此类算法既不需要考虑总体的分布,又能够保证样本的代表性。实证分析结果表明,基于此算法抽取样本构建的模型与简单随机抽样方法相比,模型参数估计的均方误差更低,效果更稳定。

Most of the existing unequal probability sampling methods for big data are based on data distribution and have poor generalization ability.An unequal probability sampling algorithm for massive data was proposed by estimating the relative distances between population elements using a multi-layer perceptron,XGBoost and Kriging models.This kind of algorithm dose not need to consider the distribution of the population,but also can ensure the representative of the sample.The empirical analysis results show that the mean square error of model parameter estimation based on this algorithm is lower than that of simple random sampling method,and its effect is more stable.