Date of Award
Master of Science (MS)
Dr. Zhengxin Chen
Dr. Yong Shi
Two important issues related to time series data are similarity analysis and cluster analysis. There are two different, but related issues. Regarding to similarity analysis, although R*-Tree based method is most promising, its performance suffers from the socalled "dimensionality curse" and thus dimensionality reduction is needed for it to function efficiently. In this thesis, we use PCA as a dimensionality reduction method in similarity analysis of time series. A similarity tool is developed with dimensionality reduction modules PCA, DFT and PAA included. Compared with DFT and PAA, PCA demonstrates better distance conservation property after dimensionality reduction, cheaper query time and post-processing time, and less false positives for both exact queries and similar queries. Furthermore based on its feature of indexing MBBs according to spatial proximity, we extend R*-Tree's application to cluster analysis. With the aid of R *-Tree indexing, we propose two clustering methods, KMeans-R and Hierarchy-R, as an improved version of K-Means and Hierarchical Clustering, respectively. The performance of two clustering methods is compared against K-Means and K-Means with sampling technique (KMeans-S). We utilize Rand Index (RI), Adjusted Rand Index (ARI) and Information Gain (IG) as the measure of clustering quality to evaluate the four clustering methods. Compared with K-Means and KMeans-S, the clustering results show that KMeans-R and Hierarchy-R can achieve better clustering quality.
Pi, Jiaxiong, "Similarity and cluster analysis of time series data using R*-trees." (2005). Student Work. 3556.
Files over 3MB may be slow to open. For best results, right-click and select "save as..."