Date of Award


Document Type


Degree Name

Doctor of Philosophy (PhD)


Computer Science

Committee Chair

Heggere S. Ranganath

Committee Member

Mary Ellen Weisskopf

Committee Member

Huaming Zhang

Committee Member

Ramazan S. Aygun

Committee Member

Peter Slater


Piecewise linear topology., Time-series analysis.


The primary contribution of this dissertation is the development of a new time series representation technique called the Hierarchical Piecewise Linear Ap- proximation (HPLA), which facilitates the development of an efficient and effective approach for time series matching. In the HPLA representation, a time series is partitioned into segments between identifiable perceptually important points called primary breakpoints, and each segment is represented independently at several levels of accuracy by placing additional secondary breakpoints between adjacent primary breakpoints. For a given compression ratio, the HPLA representation achieves higher reconstruction accuracy and retains local evolution trends better than the existing time series representation techniques. The structure of the HPLA representation en- ables the automatic selection of the sequence-to-sequence, sequence-to-subsequence, or subsequence-to-subsequence matching by aligning the primary breakpoints of the two time series being matched. The feature vectors of the HPLA representation, be- ing invariant to translation and scale, are able to determine similarity between two time series even when the two time series differ in translation and scale. The HPLA based approach adapts easily online implementation for the real-time processing of the streaming data. On the surface, the research presented in this dissertation appears to have focused and succeeded in developing a viable solution to time series matching, in- cluding the challenging subsequence-to-subsequence matching problem. In reality, the research has produced a framework and a segment-wise processing methodology that is equally suitable for the development of solutions to many important time series applications. This is because, unlike the other representation techniques which ap- pear to have been developed for specific applications, the development of the HPLA is guided by the characteristics of the ideal representation derived from the requirements of many applications. Therefore, the HPLA provides a suitable platform on which al- gorithms can be developed for several applications including clustering, classification, query by content and motif detection.



To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.