基于函数主成分的函数型数据分类研究

不同属性特征可以反映出数据不同的内在信息,越多的差异性特征对机器识别就更有利,但是越多的特征数目引起数据更高复杂度。针对函数型数据最主要的函数性和导数性这两大特征,本文提出对函数型数据函数特征、一阶导数特征和二阶导数特征的组合集成方法,然后引入函数型主成分分析的方法解决数据的复杂性问题,最后通过函数型主成分距离度量方式,采用k近邻(knn)分类以达到分类的效果。实验分析表明了函数型主成分分析方法与混合多特征组合距离的结合,在函数型数据分类中的有效性。

Functional Data Classification based on Function Principal Component

Different attribute characteristics reflect different intrinsic information of data. The more different features, the more favorable for machine recognition. On the other hand, more feature numbers cause the higher complexity of data. According to the two main features of functional data, that is functional and derivative property. This paper proposes a combined method of functional data with function, first and second derivative property.And then it introduces functional principal component analysis(FPCA) to treatthe complexity of the data. Finally k-nearest neighbor(knn) is used to achieve the classification by functional principal component distance metric. The experiment shows the effectiveness of combination of functional principal component analysis(FPCA) withmixed Multi-distance Metricsto functional data classification.