非线性面板数据聚类方法研究

对于一类变量非线性相关的面板数据,现有的基于线性算法的面板数据聚类方法并不能准确地度量样本间的相似性,且聚类结果的可解释性低。综合考虑变量非线性相关问题及聚类结果可解释性问题,提出一种非线性面板数据的聚类方法,通过非线性核主成分算法实现对样本相似性的测度,并基于混合高斯模型进行样本概率聚类,实证表明该方法的有效性及其对聚类结果的可解释性有所提高。

The Research on Non-Linear Panel Data Clustering

According to the existing research,the current clustering algorithm based on linear technique can hardly estimate the similarity between Non-Linear Panel Data and shows little interpretability of the cluste-ring result.Considering the nonlinear data structure and the interpretability of clustering result,this paper proposes a new method which characterizes the similarity of different samples using Kernel Principal Com-ponent Algorithm and applies Gaussian Mixture Model Algorithm in clustering samples.The experimental result demonstrates the validity of the method proposed above in clustering Nonlinear Panel Data and the improvement in explaining clustering result.