應用資料探勘技術建構老人慢性病人流失預測模型;Application of Data Mining Technology to Construct the Predicting Model of Elderly Patients Loss with Chronic Diseases

National Chung Cheng University Institutional Repository > 管理學院 > 資訊管理系醫療資訊管理研究所 > 學位論文 > Item A095B0000Q/492

Please use this identifier to cite or link to this item: http://ccur.lib.ccu.edu.tw/handle/A095B0000Q/492

Title:	應用資料探勘技術建構老人慢性病人流失預測模型;Application of Data Mining Technology to Construct the Predicting Model of Elderly Patients Loss with Chronic Diseases
Authors:	汪妃芳;WANG, FEI-FANG
Contributors:	資訊管理系醫療資訊管理研究所
Keywords:	資料探勘;病人流失;慢性病;監督式學習;Data mining;Patient loss;chronic disease;supervised learning
Date:	2017
Issue Date:	2019-07-17
Publisher:	資訊管理系醫療資訊管理研究所
Abstract:	高齡化之趨勢，慢性疾病的盛行率高，在競爭激烈的醫療市場環境，如何深化顧客關係管理，提升醫療服務品質，病人的忠誠度，減少顧客流失是目前醫療院所的永續經營之道，而資料探勘已是廣泛使用探討顧客流失的研究方法及技術。本研究以南部某區域教學醫院65歲以上開立慢性病處方箋病人，總共收集28,261筆就醫資料，採資料探勘分類技術之決策樹J48 (C4.5)、隨機森林，羅吉斯迴歸及支援向量機四種分類器建立65歲以上開立慢性病處方箋病人流失預測模式，實驗資料再將10組訓練樣本分別進行資料分析。隨後採用10疊交叉驗證法(10-fold cross-validation)驗證預測模式效能。實驗結果顯示以隨機森林分類技術所建構的預測模型最佳，AUC：91.33%，其次為邏輯斯迴歸，AUC：90.12%。本研究中預測因素包括最近一次門診天數、全年門診次數、門診業務量及全年門診費用是預測流失有高度相關性。期能經由本研究結果供醫院管理者針對可能造成流失之病人能即早介入預防措施，進而減少病人流失，建構完善顧客關係管理。 Currently, the trend of aging and the prevalence rate of chronic diseases has become higher. As a result, how to deepen Customer relationship management, improve the quality of medical services, patient loyalty and reduce customer loss in the highly competitive medical market environment is the way of sustainable development to current medical institutions. Moreover, the research methods and technology of data mining has been widely used to explore customer loss. In this study, a total of 28,261 medical information was collected from the teaching hospitals in the southern part of Taiwan in the area of 65 years old and adopting four kinds of classifier—the Decision Tree J48 (C4.5), Random Forest, Logistic Regression and Support Vector Machine to establish the establishment of chronic disease prescription over 65 years of age patient loss prediction model. And then, use 10 groups of training samples to analyze data. The 10-fold cross-validation was then used to verify the predictive mode performance. The experimental results show that the prediction model constructed by random forest classification is the best—AUC: 91.33%, followed by logistic regression, —AUC: 90.12%. The predictors of this study included a high degree of correlation with the number of outpatient clinics, the number of outpatient visits, the outpatient service volume and the annual outpatient cost. I expect that the results of the study can be used to help the hospital managers to intervene early in preventive measures, thereby reducing the loss of patients and improve the customer relationship management.
Appears in Collections:	[資訊管理系醫療資訊管理研究所] 學位論文

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	405	View/Open

Loading...