- 學位論文
以資料探勘方法建立病患門診決策系統
The Constructing of an Outpatient Hospital Registration Decision System Using Data Mining Methods
摘要
在現今門診科別繁多與醫療資源豐富的社會當中，民眾必須對於醫療方面的知識有所了解，才能前往正確的門診科別，但因為一般民眾對於醫療知識是不足的，所以常常為了該掛哪一種科別門診而困擾，造成了掛號錯誤的狀況，浪費了自身的時間也浪費的許多的醫療資源。 本研究為了產生門診科別建議，故以資料探勘方法來建立病患門診科別決策系統，其中包含單一分類器(ANN、SVM、CART、NBC與Logistic)決定門診科別；再以兩分類器結合(其中模型為ANN、SVM結合CART、NBC與Logistic)來進行兩階段的分類。單一分類器是將病徵輸入各分類器當中，產生門診科別，而兩階段分類是將病癥分別輸入至ANN與SVM中，產生類別門診(內科、外科、其他)縮小分類範圍，再將ANN、SVM產生之分類資料，輸入至CART、NBC與Logistic。 經組合模式實驗顯示，SVM & Logistic準確度為最高，平均準確度達96.27%，各組合模式與單一模式SVM相比較，SVM & Logistic整體準確率雖然較低，但若以類別門診分開來看，外科與其他類別門診準確率可達98.8%，比單一SVM的準確度高出1.51%，而內科準確度則較低為96.27%，建議內科門診可使用單一SVM模式來做決策。經修正，最後模式的準確度內科為97.44%、外科98.31%，其他99.32%，平均準確度為98.36%，高於單一SVM分類器1.07%。
並列摘要
Nowadays, there are many outpatient divisions and medical resources, people have to understand the medical knowledge, before they go to the right outpatients. But people’s medical knowledge is not enough, so which outpatients should be registered that always causing problems for people, caused the error condition of register also waste their own time and a lot of the medical resources. In this study, it aims to produce recommendations about outpatient divisions. Use of data mining methods to build decision-making system of disease clinic divisions, which includes a single classifier (ANN, SVM, CART, NBC, and Logistic) decided to out-patient divisions; then the two classification combination (of which model ANN, SVM combined with CART, NBC and Logistic) for a two-stage classification. Single classifier is to enter the symptoms of the classifier, resulting in out-patient division, while the two-stage classification of disease enters to the ANN and SVM respectively. The generated class out-patient (medicine, surgery, other) narrows down the range of categories. Next, the ANN, SVM classification of data will be entered to the CART, NBC and Logistic. The combination of model experiments showed that, SVM & Logistic have the highest accuracy, average accuracy of 96.27%, the combined model compared with the single mode of SVM. SVM & Logistic although the overall accuracy low, but if the patient separately, by category, other types of outpatient surgery and 98.8% accuracy rate, higher accuracy than single SVM from 1.51%, while the medical accuracy is 96.27% lower. It is proposed that the medical clinic can be made decisions by using a single SVM model. Amended, the final model is 97.44% accuracy of medical, surgical 98.31%, other 99.32%, with an average accuracy of 98.36%, higher than the single SVM classifier 1.07%.