摘要
现实生活中广泛存在的高维时序数据常常具有决策属性且时间长度不等的特点,使得现有的邻域粗糙集特征选择算法不再适用或分类性能下降。为了解决该问题,提出了一种基于DTW M度量的高维时序数据的特征选择方法。首先,引入马氏距离定义高维动态时间扭曲距离(DTW M)用于度量属性间的相似性;接着定义了时序决策信息系统,用于存放非等长高维时序数据;提出基于DTW M距离度量的时序邻域关系和时序邻域粗糙集模型;最后通过定义内、外重要度,给出了属性依赖度作为筛选和提出属性的重要指标,进而提出了基于DTW M度量的高维时序数据的特征选择方法。通过五个公开数据集上的实验结果表明,所提算法较其他算法在分类精度上平均提升了14.2%和21.7%,充分证明了其有效性和优越性。
High-dimensional time series data exist widely in real life.They often have decision attributes and vary in time length.These characteristics render existing neighborhood rough set feature selection algorithms inapplicable or reducing their classification performance.Thus,this paper proposed a feature selection method for high-dimensional time series data based on metrics.Firstly,it introduced the Mahalanobis distance and defined the dynamic time warping of Mahalanobis(DTW M)to measure the similarity between attributes.Then,it also defined a time series decision information system to store non-equal-length high-dimensional time series data.It also proposed time series neighborhood relationship and a time series neighborhood rough set model based on DTW M distance measurement.Finally,it defined internal and external importance,and presented the attribute dependency served as a key indicator for screening and selecting attributes.Thereby this paper put forward a feature selection method for high-dimensional time series data based on DTW M measurement.Experiments on five public datasets verify the method has an average improvement of 14.2%and 21.7%in classification accuracy.These results fully validate the effectiveness and superiority of the proposed method.
作者
杨璇
王潇婉
胡灵芝
吴迪
Yang Xuan;Wang Xiaowan;Hu Lingzhi;Wu Di(School of Basic Medical Sciences,Shaanxi University of Chinese Medicine,Xianyang Shaanxi 712046,China;School of Sciences,Chang’an University,Xi’an 710064,China;School of Software,Tsinghua University,Beijing 100084,China)
出处
《计算机应用研究》
北大核心
2026年第1期170-177,共8页
Application Research of Computers
基金
国家优秀青年科学基金资助项目(61822111)
陕西省教育教学改革研究项目(23ZY022)
陕西中医药大学教育教学改革研究项目(23jg01)。
关键词
特征选择
高维时序数据
DTW
M度量
马氏距离
邻域粗糙集
feature selection
high-dimensional time-series data
DTW M metric
Mahalanobis distance
neighborhood rough set
