基于图形分析方法的函数型数据异常值检验实证研究

An Empirical Study on Outliers Test of Functional Data Based on Graph Description Tools

函数型数据本质上是一种复杂数据,其抽样、生成、结构和关联程度都会影响到数据的复杂性和描述性,有些情形甚至连基本的可视化描述都成为难点。在利用函数型数据的主成分得分、图基的数据深度和密度概念的基础上,引入函数型数据的打包图和箱线图,并针对函数型数据的图形分析提出了函数型数据异常值检测的三种方法。与已有的检测方法相比较,所提三种方法更易于识别函数型数据的异常值。

As one kind of complex data essentially ,functional data can be affected in its complexity and descriptive nature by sampling ,producing ,structure and degree of association easily ,even basic visual descriptions of it has become difficult .In general ,the usual graphical -analytical method of functional data is the traditional method called data smoothing .Through this method ,status ,structures and trends of functional data can be described by the smooth curve .This paper introduces package diagram and box plot of functional data into the scores of principal component and the concept of data depth and density of Turkey . Simultaneously , while graphical analysis of functional data develops , tests of outliers in functional data have been resulted .Compared with the existing detection methods ,the proposed method by this paper is easier to identify the outliers of functional data .