透過您的圖書館登入
IP:3.144.151.106
  • 學位論文

以監督式學習方法進行檢驗管控

Quality Control by Supervised Learning Method

指導教授 : 周珮婷 林怡伶

摘要


本研究之動機為將探討傳統的進料檢驗管控(Incoming Quality Control, IQC)之允收抽樣計畫之假設、特性以及允收過程,將其關鍵想法做為資料與變數模擬之依據,並藉由該模擬資料進行監督式機器學習模型之配適,預測材料或零件供應商所提供之抽驗資料是否具有造假之意圖。 首先,本研究依照允收抽驗計畫的假設特性,將利用供應商抽到未符合標準公差之抽樣零件時即進行重新抽取樣本直至符合其標準的行為視為造假資料,並使用遞迴的方式進行模擬。再來,運用支持向量機、羅吉斯迴歸以及隨機森林等監督式學習方法進行預測,並比較各個變數的預測效果。 從結果來看,依照允收抽驗樣本選擇的變數對於分辨供應商資料是否造假具有不錯的效果,依照本研究之結論,企業可依照供應商之抽驗資料轉換特性並建置供應商管理判別系統,並利用該方式作為供應商的選擇以及評估,其必可降低企業之鑑定成本(Appraisal Cost) ,造就供應商、零售商與客戶之間的三贏局面。

並列摘要


The purpose of the current study was to explore the assumptions, features, and acceptance process of acceptance sampling plan in traditional Incoming Quality Control (IQC). Four features were proposed to describe distributions of data. Supervised machine learning models, Support Vector Machine(SVM), Logistic Regression, and Random Forest, were applied for detection of fraud. The results showed that the proposed features can effectively differentiate between real and fake datasets. The techniques can be used in future for supplier selection and evaluation. The identification of appraisal cost will be reduced and a triple-win situation for suppliers, retailers, and customers can be created.

參考文獻


參考文獻
Alpaydin, E. (2010). Introduction to machine learning (2nd ed.). Cambridge, MA: MIT Press.
Boulesteix A-L, Tutz G. (2006). Identification of interaction patterns and classification with applications to microarray data, Comput. Stat. Data Anal.
Breiman, L. (2001). Random Forests. Machine Learning, 45(1), 5-32.
doi:10.1023/a:1010933404324

延伸閱讀