近年來，線上拍賣的蓬勃發展有目共睹。線上拍賣交易兼具便利性與隱蔽性，且不受時間與空間的限制，使得交易量逐年顯著提升。然而，面對如此蓬勃的交易平台，許多詐騙者開始混雜其中，謀取不法利益。詐騙的方式不但多樣化，且經常隨著時間、環境改變，令人防不勝防。為了協助交易者早期發現詐騙陷阱，避免蒙受不必要的損失，本研究以行為狀態分析為基礎，發展了一套線上拍賣詐騙偵測與預警方法。首先，針對詐騙者及正常者的交易記錄進行時序切割，再對其特徵值向量進行分群，以歸納出典型的交易者狀態。而後，針對資料集中所有的交易歷史進行狀態變遷切割，以產生與時序行為相關的偵測模型。在此同時，我們也利用狀態切割後的資料集，製作狀態標籤字串，並產生循序樣式，供使用者比對、監控可疑帳號。根據上述方法，本研究實作了一套簡易的線上拍賣交易輔助系統，輔助使用者在交易前觀察、分析交易對象的行為。為了驗證提出方法之有效性，本研究使用拍賣網站實際交易資料進行實驗，結果顯示本研究提出之方法確實有助於提升詐騙偵測之準確性與預警能力。

Purpose－The fraudsters’ strategies of online auctions are diverse and changing rapidly. It results in the difficulty of fraud detection and prevention. The purpose of this paper is to develop effective methods to help discovering online auction fraud as early as possible. Design/methodology/approach－This paper develops effective detection methods based on behavioral state transition of fraudsters. First, we partition and duplicate the transaction histories of traders according to trading events. Then, a reduction method based on state transition is developed to reduce the size of data set, which is then used to build the detection model. In addition, the state label strings are used to conduct the behavioral patterns of suspects for monitoring. Findings－To demonstrate the effectiveness of the proposed methods, real transaction data are gathered from online auction sites for experiments. The results show that our methods do increase the detection accuracy and demonstrate that the early fraud detection by behavioral monitoring is possible. Research limitations/implications－The limitations of this work is that the proposed method could be ineffective for the fraudsters who steal or buy other normal accounts for disguise. Albeit being difficult, it is still possible to discover them by monitoring their behavioral changes in some critical time point. Certainly, it needs newly-developed detection methods. Practical implications－If the developed methods can be implemented and incorporated into the routine tasks of real online auction sites, the efforts of monitoring abnormal traders can be greatly reduced and the cost of maintaining a smooth trading environment can drop significantly. As a result, the fraud events will be effectively suppressing and the users will have more confidence in trading with online auctions. Originality/value－To apply state transition concept to detect latent fraudsters, which extends intuitive decision tree and other learning models to more complicated time-based analysis. Thus, based on the proposed novel approaches, new methods can be developed to discover more well-camouflaged fraudsters.