現今醫學之發達,但對於胸痛形成的病因無法立即判斷,引起胸痛的原因非單一原因,此時必須依靠醫生長年的臨床經驗來判斷。在判斷胸痛的主因中,病歷資料可成為一個判斷的方向,但在現實中,醫生不可能在有限的看診時間內瀏覽完,尤其是在緊急的情況下(例如急診),因此一個有效的醫療資訊輔助系統是需要的。在這篇研究中,我們著重在建立胸痛的分類模組,由病人的過去歷史資料來協助醫師判斷此病患是否患有胸痛。我們使用了決策樹、向量學習機器、貝氏網路以及樸素貝氏網路等資料探勘的方法來建立我們的分類模組。本篇論文的匿名病歷資料是來自於台灣嘉義基督教醫院101-102年的急診資料,經過資料前處理的步驟後,有效的總資料數為103,893筆,其中胸痛患者有4,139筆。藉由實驗結果來看,貝氏網路是五個資料探勘方法中最適合胸痛的分類模組,其正確率和捕捉率皆達到90%以上。
To provide an appropriate diagnosis, physicians need to have overall comprehension of a patient's health information, which is usually a huge amount of data that can only be reviewed partially under tight time constraints. An effective way for physicians to determine the digested critical medical information of a patient is required. In this thesis, we focus on mining useful patterns for building chest pains classification models based on medical records. We use several existing methods, including the Decision Tree, Support Vector Machine, Bayesian Network and Naive Bayes, to build our classification models. The dataset comes from the emergency department of Chiayi Christian Hospital, Chiayi City, Taiwan. The time interval of the dataset is from 2012 to 2013. There are 4,139 patients who suffer from chest pain among all 103,893 patients after the data preprocessing step. From the experimental results, our models have high accuracy. The BN classifier performs the best, which is more proper to the chest pain dataset than other four classifiers.