本研究以更有效的利用原始資料之方法-資料修補機制來探討在醫療環境中可能發生的資料缺失問題，在此提出了一套資料修補機制。首先，應用原本存在於資料中隱含的資訊-特徵項目間之關連性，並且結合了專家知識，使其建立特徵項目之修補架構。再者以類神經網路當中效果良好的一項工具-函式模擬，用以達到補上缺失特徵項目之值。最後應用一種在醫療環境當中常見的現象-少量資料、特徵項目多資料當中，分類效果較為優異的理論方法-支援向量機器，來判斷檢驗本研究之修補機制的結果。 本研究以兩種類型之實例試驗來評估資料修補機制的可行性。第一類型為本身資料結構為完整的資料集型態，目的是方便檢驗資料修補機之成效。以人為的方式使其資料集內容由完整而成為有缺失的資料欄位。並且為了模擬資料缺失的狀態，分別模擬了高度資料缺失、中度資料缺失、以及少量資料缺失三種情境，希望更能切合資料缺失的現實狀態。第兩類型資料便是實際與醫院合作收集之醫療資料，用來做為資料修補機制之佐證。 在本研究進行的實例驗證當中，人為所製造之缺失資料集用資料修補機制來修補其缺失資料空格。以缺失資料跟修補過後之資料用以支援向量機器來做為分類測試，其結果皆為顯著。說明了應用本機制後，對於資料的正確判斷有明顯的增加。 The incomplete-data is the critical problem in the medical environment. In this paper, we provide the data-repairing technique in order to get more information from the primitive data. First, the application exists originally in data in implicit information- relation of the features, and combined expert''s domain knowledge, making it create the features to repair the structure. Using the method that Function Approximation is the application in the foundation of Neural Network to fill block of incomplete-dataset up. And the one of excellent method in the classification - support vector machine, judges examination this research it repairs the mechanism of result. This research evaluates the feasibility that data-repairing mechanism is with two of the category types. The first type is oneself the type of data set that data structure is complete; the purpose is a convenience examination data to repair the result of mechanism. Make by factitiousness that it’s the content of data set and become the data field of having the missing. For simulating the status of missing-data, simulated the mass incomplete-data, medium degree incomplete -data respectively, and a little incomplete-data. There are three kinds of scenarios, hoping can also suit realistic status of the incomplete-dataset. The second type data would be to in concert with the medical data of the collections with the hospital physically, using to be used as the data to be the substantial evidence of data-repairing mechanism. Identification that this research, the incomplete dataset that factitiousness makes repairs the mechanism to fill its blank space of incomplete-data with the data-repairing mechanism. Repair with the incomplete-data later on its data supports the vector machine in order to be used as the classification, the result is obvious. After explaining application this mechanism, have the obvious increment for the right judgment of the data.
