在現實生活中，常常會面對許多不完全的資料，而約略集合理論(rough set theory)採用某種衡量標準來分割資料，是常被使用的處理方法。但是大部份情況下，約略集合理論均使用離散型資料(discrete data)來推導規則，而實際上資料中常會包含數值型資料(numeric data)，並且會有許多不必要的屬性。因此針對此類問題的發生，本論文建構一個完整資料挖掘架構對此類資料作前處理，並使用可變精確度約略集合理論(variable precision rough-set model;VPRSM)來產生決策規則。 此完整資料挖掘架構包含三個步驟，屬性轉換、屬性篩選及規則產生。第一個步驟是屬性的轉換，使用卡方統計量將數值型資料離散化以轉換成區間型資料。第二個步驟是屬性篩選，將不重要的屬性刪除。第三個步驟為規則產生，使用可變精確度約略集合理論來產生決策規則。在實驗中，我們以患有單一肺部結節(SPN)之病人為例，來作為實作的學習樣本。實驗的結果顯示經由本論文所建構的完整資料挖掘架構可以獲得較精確之決策規則。 In literature, the rough set theory is a widely used method to induce data into rules with meaning information based on certain measurements. The induction method of the rough set theory generally generates rules from these discrete data; however, we usually deal with a lot of numeric and incomplete data with ambiguous features in daily life. In order to overcome these problems the paper constructs a complete architecture to generate decision rules by using Variable Precision Rough Set Model (VPRSM). The complete architecture includes three steps that are feature transform, feature selection and rule generation. Feature transform uses Chi-square estimator as criterion to transfer data with numeric attributes into interval ones. Feature selection searches for powerful features for sequential rule extraction. The goal of rule generation is to generate decision rules by using VPRSM. In experiments the disease history records of SPN are collected to explore the performance compared to conventional methods. The results demonstrate that the better discriminate results have been obtained with respect to the conventional ones.
