線上拍賣蘊含龐大商機，但詐騙者也開始混雜其中，讓消費者防不勝防。面對日益猖獗的線上拍賣詐騙，除了提醒交易者小心謹慎外，學者們提出各種詐騙偵測方法。一般而言，詐騙偵測的準確性與分類屬性集的效能息息相關。然而，前人大多使用經驗法則來設計屬性集，我們認為應有更系統化、更周全的考量。有鑒於此，本研究致力於發展詐騙偵測屬性集的挑選與建構方法，以提升詐騙偵測的準確性。為達成上述目標，首先，我們提出了一套基因式的屬性挑選方法，並設計了一套完備適應函數。在演化過程中，除了偵測準確率外，也同時顧及偵測成本的多寡，期能產生一組低成本、高效能的詐騙偵測屬性集。接著，本研究發展了一套語法演化式的屬性建構方法，以BNF為基礎，配合基因演算法，以各種不同方式組合原生屬性，以產生高效能的複合屬性。為了驗證提出方法的有效性，我們使用拍賣網站真實交易資料來進行實驗。實驗結果顯示，針對不同資料集，本研究提出的方法能有效縮減屬性集的大小，並獲得較佳的準確率。此外，語法演化後產生的新屬性也具有良好的偵測成功率，有助於總體準確的提升，與資料集維度的縮減。 Because of big commercial opportunity in online auctions, there are more and more fraudulent incidents. It is also difficult to let consumers aware fraudulent transactions. In the face of fraud in online auction, many scholars have proposed some fraud detection methods instead of reminding consumers to be careful. Generally, the success rate of fraud detection has a big relationship with the fraud detection feature set. Most of scholars designed their own feature set depends on experiences. In order to improving success rate of fraud detection and generating our feature set automatically by system. In this paper, we propose a BNF-based grammatical evolution method in feature construction and a genetic algorithms in feature selection for fraud detection. The grammatical evolution technique inspired by natural evolution is explored to detect fraudsters in online auctions. Moreover, we illustrate the effectiveness of our algorithm on a real dataset collected from a large online auction site Yahoo.
