基于主成分分析和Eros的邻近传播算法在金融数据集中的应用

Application of PCA and Eros Affinity Propagation Clustering in Financial Data Sets

金融数据集的多维特性和高噪声特性使得对金融时间序列数据的分析难上加难,本文提出一种基于主成分分析和Eros的近邻传播的聚类算法。首先利用主成分分析方法对多变量的金融时间序列数据进行降维处理,提取出主要特征值;然后使用基于Eros的近邻传播算法聚类对提取出的特征值进行分析。该聚类方法可以把数据集中的个体当作是原始数据的一个属性,通过迭代竞争达到最优,不需要事先确定聚类数目。研究结果表明,这种集成算法大大降低了时间序列数据的维度,有很高的分类正确率,表明该聚类方法用于金融时间序列数据处理是有效可行的。

The muhi-dimensional characteristics and high noise characteristics of the financial data set make it hard to analyze the time series. This paper puts forward an algorithm based on the principal component analysis and the Eros affinity propagation clustering. First it uses the principal component analysis method to extract the main eigenvalues of the muhivariate financial time series data; then uses the Eros affinity propagation clustering to analyze the extracted eigenvalues. This kind of clustering algorithm can make the individual data as an attribute of the original data, through iterate competition to achieve optimal, do not need to find the number of clusters. The research results show that, this integrated method greatly reduces the dimension of the time seties, and has a highly correct classification rate. It proves that this algorithm is very effective.