本研究主要是在探討如何在時間序列資料上探勘部份週期性樣式。在以往大多數的研究中,都採用了Apriori [18] 演算法中的特性來探勘部份週期性樣式。然而,此方法將會產生過多的暫時性中間資料庫而導致時間和空間上的浪費。另外,於真實世界中不同事件的發生頻率是不相同的。因此,只採用單一門檻值是不夠的,它無法反映出在真實世界的現象。 在本文中,我們提出了一種修改至 FP-tree 結構的週期性樹狀結構 (我們稱為 PFP-tree)。另外,我們將單一門檻值延伸至允許使用者自定的多重最小支持度,來反映真實世界現象。最後,我們將利用實驗去證明我們的研究是有效的。本研究可應用於股票市場的活動、天然災害的預測、人們的購物習慣等等。例如:在線上購物系統的資料庫中,我們可以透過客戶的登錄時間和瀏覽的商品來得知一些具有週期性的資訊,從這些週期性行為中可獲得一些有趣的資訊,並且利用它們來提升企業在做預測和決策時的精確性。
In this study, we have studied the problem of mining partial periodicity in time series database. Most of the previous studies adopt an Apriori-property to mining periodic patterns. It is based on an Apriori heuristic [18]. However, the time cost of candidate set generation is expensive. On the other hand, using a single minimum support can not reflect the real-life situation. For this reason, we propose a periodicity tree (PFP-tree for short) structure. It is design by modifying the FP-tree structure. Moreover, we also developed an efficient algorithm to mining periodic patterns with multiple minimum supports. On the other hand, we demonstrate the usefulness of these techniques through an extensive experimental study. Our research can be applied to stock market price movement, natural calamities prediction (e.g., earthquake) and a person’s shopping habit, etc. For example, in a database maintained by online shopping system, we can get periodic information of customer's shopping habit through login time and products of searching for each user. Their periodicities may reveal interesting information that can be used for prediction and decision making.