论文部分内容阅读
作为一种重要的数据类型,函数型数据经常出现在实际应用问题当中.当输入是函数型数据输出是类别标签时,该实际问题就变成了函数型数据的分类问题.函数型数据通常具有高维、自相关等特点,抽取关键趋势特征是对函数型数据两阶段机器学习的重要一环.一方面可以避免维数灾难,另一方面可以保留重要的判别特征.函数主成分分析是一种由数据驱动的、对函数型数据进行降维处理的有效方法.然而,离群函数样例和样例间特征未对齐等因素使得函数主成分对函数型数据的表示能力退化.为此,本文提供了一种对函数型数据进行纵向标准化变换的方法,即将每个函数样例的值域变换到单位区间且不改变函数样例的整体变化趋势的一种变换,并指出该变换能够提供较为稳健的函数主成分并为改善分类精度奠定基础.
As an important data type, functional data often appear in the practical application of the problem.When the input is the output of functional data is a category label, the practical problem becomes the classification of functional data.Functional data usually has High dimension and autocorrelation, extracting the key trend features is an important part of the two-stage machine learning of functional data, which can avoid the dimensionality catastrophe on the one hand and retain the important discriminant features on the other hand.The principal component analysis However, the exponential function of the principal component of the function degrades the representation ability of the functional data.Therefore, This paper presents a method for the longitudinal standardized transformation of functional data by transforming the range of each function sample into unit intervals without changing the overall trend of the function sample and pointing out that the transformation can provide The more robust principal components of the function and the basis for improving the classification accuracy.