結合模糊集合理論與貝氏分類法之資料探勘技術--應用於健保局醫療費用審查作業

A Data Mining Technique Combining Fuzzy Sets Theory and Bayesian Classifier - An Application of Auditing the Health Insurance Fee for the National Health Insurance

藍中賢(Chung-Hsien Lan)

指導教授 ： 詹前隆

元智大學/資訊學院/資訊管理學系/碩士(2000年)

摘要

資訊科技的出現，早期主要是以龐大的計算能力，來協助人們在工程和科學方面的需求。然而，隨著科技的迅速發展，資訊技術逐漸被廣泛應用到組織當中，從傳統的電子資料處理，到決策支援系統，甚至到現今廣受企業重視的資料探勘〈Data Mining〉和資料倉儲〈Data Warehouse〉技術。為了有效分析過去的資料，建立解決問題的規則，必須採行資料探勘的技術，以期節省更多的資料分析成本與時間。
資料探勘技術中有許多的技術都可用來協助分類的問題，特別是貝氏分類法，因為貝氏分類法的核心理論是貝氏推論，所以在進行資料分析時，能考量所有影響分類結果的因素，並且能清楚判定案例所屬類別，而且對於推得的結果能有明確的解釋。但是，貝氏推論在處理連續的屬性值方面，其計算方式十分複雜，必須考量不同分配的整合方式。因而，本研究藉由模糊集合理論，來建立連續的屬性值的區隔，將連續的屬性值轉成離散的屬性值，然後，結合貝氏分類法建立資料探勘與決策之模式。
利用模糊貝氏分類法建立的雛型系統落實資料探勘的目的，確實將資料經過處理或解釋轉為資訊，並將資訊運用於決策或驗證，然後累積形成知識。一旦採行此模糊貝氏分類法的技術，便能有效分析大量資料，找出隱含的規則，以協助往後的推測工作。
在應用方面，引入醫療保險費用審查之資料，並由分析結果中得到敏感度（sensitivity）為0.639，鑑別率（specificity）為0.968，因而得知系統分析成效良好。如此一來，便能利用此系統協助醫療費用的審查作業，進而管理與監控醫療保險費用的成長。

關鍵字

資料探勘 ； 模糊集合 ； 貝氏分類法 ； 醫療保險費用 ； 異常偵測

並列摘要

Information technology is widely applied in business from traditional electronic data processing to cutting-edged decision support systems, data mining and data warehouse today. To analyze historical data effectively and to build the problem solving rules, data mining techniques are used to reduce the cost and time for data analysis.
Many data mining techniques can be used to solve “classification” problems, especially the Bayesian classifier. Bayesian classifier based on Bayesian inference, because it can take all attributes affecting the classification result into account. In addition, the Bayesian classifier is used to judge the class of a case clearly and explain the result in detail. However, when dealing with continuous attributes, Bayesian inference needs to integrate different probability distributions. In this case, the computation is very complex. Therefore, the research transforms the continuous attributes with the Fuzzy sets theory by converting continuous attributes to discrete. Consequently, a new data mining models by incorporating the Bayesian classifier and the Fuzzy Set Theory can be constructed. The prototype based on the Fuzzy Bayesian classifier fulfills the purpose of data mining technique --- transform data to information, and transform validated information into knowledge. Hopefully, Fuzzy Bayesian classifier can be used to analyze huge amount of data and to find the hidden rules effectively.
For the application, Fuzzy Bayesian classifier is used to analyze a portion of health insurance fee data for validating the decision model granting the application fee. In this case, the sensitivity is 0.639 and the specificity is 0.968. Therefore, the effectiveness of this model is supported. In the end, this system can be applied to audit the health insurance fee and control the increasing speed of health insurance fee.

並列關鍵字

Data Mining ； Fuzzy Sets ； Bayesian Classifier ； Health Insurance Fee ； Fault Detection