識別序列數據中的變化，稱為改變點偵測，已成為各個領域越來越重要的話 題。改變點偵測法可以分為即時和線下，我們主要針對即時改變點的方法做研究與推廣，稱之為 EXact Online Bayesian Changepoint Detection (EXO)，已經對於真實資料顯示出合理的結果。其中，對於資料型態，EXO 假設資料點間是相互獨立的，在真實資料中，資料間其實是有一定的相關性的，對於這種有相關性的資料，我們使用 Clayton copula 之下的馬可夫鏈模型，邊際分配的部分我們使用卜瓦松去描述這種間斷型的資料。從模擬得知在強相關性的情況下，這個模型有較好的準確性。並在實證資料中這個模型與 EXO 方法得到相同的結果。;Detecting the structure change in sequential data, known as changepoint detection,has become increasingly important in various fields. As the changepoint detection method can be categorized by online and offline, this research focuses on the online way called EXact Online Bayesian Changepoint Detection (EXO). However, EXO assumes that the datapoints are independent of each other, but this may be unrealistic. For real data, there is a certain relation between the datapoints. Therefore we consider the Markov chain model under the Clayton copula with the Poisson distribution as the marginal distribu tion to describe the data with the dependence structure and illustrate the performance in simulation studies. The data analysis comes from empirical studies.
