| Summary: | 碩士 === 國立中興大學 === 資訊科學與工程學系 === 103 === Time series segmentation is one of the current research topics on data mining. As recent studies often apply time series segmentation in stocks analysis and big data processing, identifying the characteristics or turning points on the time series can effectively facilitate the analysis and forecast with the data set. Among a number of approaches to time series segmentation, PLR (Piecewise Linear Representation) has been regarded as the most classic one. It shows high accuracy, yet it takes considerable amount of time to conduct the algorithm, especially for big data segmentation. In view of such issue, this research proposes a solution which can not only improve the analysis efficiency with PLR, but also ensure that the result is within the error bound generated by PLR.
In this thesis, we propose a time series segmentation method called Piecewise Linear Representation based on Feasible Space (PLRFS). In this method, we find segmentation points based on feasible space as the time series data streaming in. The proposed method has two features. One is that it facilitates processing time with errors within an acceptable range, and the second is that the segmentation can be conducted as soon as a feasible space is located without collection of complete data. According to the experimental results, the proposed method only takes 1/12 of the processing time required by the original PLR algorithm. Moreover, the trend of original data can still be observed with the set of segment points generated by the proposed method.
|
|---|
