摘要
针对函数型数据分类算法中全局统计特征表达能力有限,且显著点特征易受噪声干扰等问题,提出一种基于统计深度方法的函数曲线特征分段提取算法。首先,利用数据平滑技术对离散观测的数据进行平滑化处理,同时引入函数型数据的一阶和二阶导函数;然后,分段计算函数本身及其低阶导函数的马氏积分深度值,在此基础上构造函数曲线特征向量;最后,给出三种选择调节参数的搜索方案,并进行分类研究。在UCR数据集上的实验表明,与当前其他曲线特征提取算法相比,所提算法能有效提取函数曲线特征,提高分类的准确性。
Since the representation ability of statistical global feature for functional data classification algorithm is limited,and the salient point feature is susceptible to noise disturbance,this paper proposed a segmental feature extraction algorithm based on statistical depth notion. Firstly,it used the smoothing technique to pre-smooth the discrete observed data,and introduced the first and second derivatives of the functional data. Then,it calculated depths of Mahalanobis integral of the functions and its low-order derivatives in segments,and thus constructed feature vectors of function curves based on the depth measures. Finally,it selected the optimal number of segments for classification by data-driven,and studied the binary classification of function data. Compared with the other curve feature extraction algorithms,experiments on UCR datasets show that the proposed algorithm performs well in extracting the feature of curve,and improves the classification accuracy effectively.
作者
金海波
马海强
Jin Haibo;Ma Haiqiang(Dept.of Mathematic,Taiyuan University of Science&Technology,Taiyuan 030024,China;School of Statistics,Jiangxi University of Finance&Economics,Nanchang 330013,China)
出处
《计算机应用研究》
CSCD
北大核心
2020年第6期1765-1768,共4页
Application Research of Computers
关键词
函数型数据
分段特征
深度函数
函数型数据分类
functional data
segmental feature
depth function
functional data classification
