案例式推理(Case-Based Reasoning, CBR)系統求解問題時,主要是利用過去求解案例中所獲得的經驗,用來推測目前求解問題結果。雖然此工具在研究上廣受使用,但是目前僅有少數研究進行改善CBR系統於數值預測研究。所以本研究發展一套新的索引方法,並且應用簡單的變數加權方式,改善CBR系統對於數值預測的準確性與效率。本研究所提出CBR系統,藉由數個UCI(University of California-Irvine)資料集合,進行預測的準確性與效率之比較。然而,本研究所提出的CBR系統也應用於求解交期指派的問題,實驗模擬於一個動態的晶圓製造工廠,並觀察系統於實務問題中能否達到預期的優勢效果。
Case-based reasoning (CBR) solves new problems by recalling and reusing specific knowledge obtained from past experiences. Despite its popularity and simplicity, little work has been done for improving CBR for numeric prediction. In order to predict numeric values accurately and efficiently, this paper typically focuses on the development of a novel case indexing approach and application of a simple attribute weighting method for CBR. The proposed CBR system is evaluated on the seven well-known data sets, exhibiting better efficiency and accuracy than the conventional CBR. This study also applies the proposed CBR system for solving the due date assignment (DDA) problem in a dynamic wafer fabrication factory in order to investigate whether it’s expected benefits can be observed in practice. The results of the experiments show that our proposed CBR system leads to substantial improvement in predicting job due dates.