时间序列挖掘聚类(PPT 61页).pptVIP

股票数据聚类：数据的归一化 针对股票数据间的股价差距大的问题，采用归一化处理，归一化处理主要解决比较数据间量纲不统一的问题，在对股票进行聚类分析中，股票的相似性集中于股价变化趋势的相似性，而非股价之间的相似性，所以采用以下公式 对数据进行归一化处理。 股票数据聚类：聚类结果 运行层次聚类算法时初始设定聚类簇数为4个，同时设定时间弯折窗口w为3。 股票数据聚类：聚类结果 运行层次聚类算法时初始设定聚类簇数为4个，同时设定时间弯折窗口w为3。 股票数据聚类：聚类结果 运行层次聚类算法时初始设定聚类簇数为4个，同时设定时间弯折窗口w为3。 股票数据聚类：聚类结果 运行层次聚类算法时初始设定聚类簇数为4个，同时设定时间弯折窗口w为3。 基于SAX表示的聚类 Hierarchical Clustering Compute pairwise distance, merge similar clusters bottom-up Compared with Euclidean, IMPACTS, and SDA 基于SAX表示的距离 PAA distance lower-bounds the Euclidean Distance 0 20 40 60 80 100 120 - 1.5 - 1 - 0.5 0 0.5 1 1.5 C Q 0 20 40 60 80 100 120 - 1.5 - 1 - 0.5 0 0.5 1 1.5 C Q = baabccbc C ? = babcacca Q ? Euclidean Distance dist() can be implemented using a table lookup. Hierarchical Clustering We can objectively state that SAX is superior, since it correctly assigns each class to its own subtree. 数据类别事先已知：decreasing trend, upward shift and normal classes Clustering Hierarchical Clustering Compute pairwise distance, merge similar clusters bottom-up Compared with Euclidean, IMPACTS, and SDA Partitional Clustering K-means Optimize the objective function by minimizing the sum of squared intra-cluster errors Compared with Raw data 比层次聚类具有更好的可伸缩性 Partitional (K-means) Clustering Working with an approximation of the data gives better results than working with the original data. It has been shown that initializing the clusters centers on a low dimension approximation of the data can improve the quality, this is what clustering with SAX implicitly does. A comparison of the k-means clustering algorithm using SAX and the raw data. The dataset was Space Shuttle telemetry, 1,000 subsequences of length 512. Surprisingly, working with the symbolic approximation produces better results than working with the original data 动态时间序列聚类 Time Series Epenthesis: Clustering Time Series Streams Requires Ignoring Some Data 动态时间序列聚类 所谓流数据，是指按照一定的时间顺序，以较快的速度连续到达的数据序列，也称为动态时间序列