序列樣式探勘是資料探勘技術中很重要的一項技術，其可經由時間的關係來瞭解各事件所發生的行為。在序列的資料中，事件可被區分為兩種主要的類型，分別為「點式事件」以及「區間式事件」，而當一序列可同時包含此兩種類型的事件時，我們稱這種序列為「混合事件」的序列。近來的研究中，吳以及陳博士所提出的演算法，可以用來找出混合事件的循序樣式，並且稱此樣式為混合樣式。雖然混合樣式的確可找出混合事件的時間先後關係，但卻無法從中瞭解各混合事件間的時間間隔。因此在本論文中，我們將延伸去探討混合樣式，來使得各事件的時間間隔可被顯露出來。我們提出兩種演算法ti-HPrefixSpan 和 ti-HTPM (分別修改自傳統的PrefixSpan和GSP演算法) 來找出具有時間間隔的混合樣式。在實驗中，我們透過模擬的資料與實際股市的資料來評估演算法的效能，並且從結果中知道我們所提出的兩種演算法都具有相當不錯的實驗成效。此外，相較於混合序列樣式，我們也從實驗中進一步分析出具有時間間隔的混合樣式確實能擁有更好預測準確率。 Sequential pattern mining can reveal the behaviors of events along the time, and it is one of the most important approaches in data mining. In sequence data, events can be classified into two major types, where one is point-based events and the other is interval-based events. When a data sequence can contain both types of events, we say this data sequence has hybrid events. Recently, the work of Wu and Chen has proposed algorithms to discover sequential patterns of hybrid events, called hybrid temporal patterns. Although hybrid temporal patterns can reveal the temporal relationships among hybrid events, they can not tell us the time intervals among events. Therefore, this thesis extends hybrid temporal patterns so that the time intervals among events can be revealed. We call this type of extended patterns as time-interval hybrid temporal patterns. In the thesis, we proposed two algorithms, ti-HPrefixSpan and ti-HTPM, to discover time-interval hybrid temporal patterns by modifying traditional PrefixSpan and GSP algorithms, respectively. In the experiments, we evaluate the proposed methods’ performance using synthetic data and real stock price data. The results of the experiments show that the performances of the two proposed algorithms are quite satisfactory. Besides, we also show that time-interval hybrid temporal patterns can obtain higher prediction accuracy than hybrid temporal patterns.
