| 摘要: | 時間序列是指每隔一定時間將所觀察的目標記錄而成的資料。故在日常生活中， 有許多現象都可以視為時間序列資料，因此，在時間序列上亦發展了許多不同的研究 議題與成果。然而，現有的演算法皆屬於資料衍生的時間序列資料探勘技術，而其主 要缺點即是探勘後的資訊皆需經過專家進一步分析後方能運用。所以，近幾年興起了 領域衍生的資料探勘概念。可惜的是，截至目前為止，所有的時間序列探勘文獻都屬 於資料衍生的探勘技術，在領域衍生的時間序列資料探勘中，仍未見有研究進行探討。 有鑑於此，根據領域衍生的資料探勘中的四種架構，本計畫預計以兩年的時間研 發可行動知識樣式的時間序列探勘技術，並針對下列的研究議題進行探討： (1) 針對單變數時間序列，根據PA-AKD 與UI-AKD 架構，研發樣式為基礎的可行 動知識樣式時間序列切割與探勘技術。 (2) 針對多變數時間序列，根據 CM-AKD 與MSCM-AKD 架構，研發樣式為基礎 的可行動知識樣式時間序列切割與探勘技術。
A time series is composed of lots of data points, each of which represents a value at a certain time. Many phenomena can be represented by time series. Many time series research topics and approaches have also been proposed. However, these time series mining algorithms are data-driven data mining approaches. And the main disadvantage of data-driven data mining approaches is that the derived patterns always need further analysis before they can be utilized. Thus, in recent years, a new concept “Domain-Driven Data Mining”, in short D3M, has been discussed. Unfortunately, all of the literature on time series data mining, to our best knowledge, is confined to the data-driven time series data mining environment; no research work has been conducted on domain-driven time series data mining. The aim of this project is to develop the AKD-pattern-based time series data mining methods, according to four frameworks of D3M. In this context, we will propose a two-year project, focusing on the following main issues: (1) For univariable time series data, according to PA-AKD and UI-AKD frameworks of D3M, we attempt to design AKD-pattern-based time series segmentation and mining techniques. (2) For multivariable time series data, according to CM-AKD and MSCM-AKD frameworks of D3M, we attempt to design AKD-pattern-based time series segmentation and mining techniques. |
