摘要
针对相似度搜索是时间序列数据挖掘的基础,构造鲁棒的动态时间弯曲距离是相似性研究的关键,考虑时间序列特征点的重要意义,引入一种时间序列波动点的抽取方法,采用二叉特征树结构对原序列进行再表达.该方法既提取了序列整体趋势信息,又有效约减了数据维数.对多个数据集的层次聚类实验表明,在保证较高准确率情况下,该方法显著提高了DTW的计算效率.
Similarity search is the foundation of data mining in time series,while constructing a robust dynamic time warping distance is the first step.Considering the importance of feature points of time series,a feature extraction method based on fluctuation is proposed,and a binary feature tree building algorithm is given.The method extracts the whole changing trend and meanwhile effectively reduces data dimensions.Clustering experiments on several datasets show that the new method is much faster and more accurate than other methods.
出处
《控制与决策》
EI
CSCD
北大核心
2007年第2期160-163,共4页
Control and Decision
基金
国家自然科学基金项目(60373107)
关键词
数据挖掘
相似度搜索
动态时间弯曲距离
特征抽取
聚类
Data mining
Similarity search
Dynamic time warping distance
Feature extraction
Clustering