本研究發展了一套具早期預警功能之線上拍賣詐騙即時偵測系統AntiFraud，當使用者以瀏覽器瀏覽商品網頁時，AntiFraud便同時啟動，進而主動判別該商品的賣方為詐騙者之可能性，以減少使用者遭受詐騙的機會。系統核心部分，整合了混合階段模型與分階段模型兩類塑模方式，用以擷取詐騙者在各階段的行為特徵來達成早期預警的功效。 為強化偵測功能，本系統建置了二層式偵測流程，第一層先判別是否為詐騙者，第二層再分辨詐騙者所處的生命週期階段。此外，本系統會根據學習結果產生詐騙類型判斷規則，提供使用者參考。本研究另外整理2007年至2009年的詐騙資料，並分析其詐騙方式演化的趨勢，使得未來可根據詐騙行為的變化提早建立防範對策。為驗證系統效能，我們使用Yahoo-Taiwan實際交易資料進行測試，結果顯示其詐騙偵測準確性與早期預警功能均具有實用價值。 This study proposes a real-time online auction fraud detection system named as AntiFraud. As a user executes Firefox to browse commodity pages, AntiFraud will be activated simultaneously to detect the auctioneers for reducing the probability of being defrauded. The kernel of the system consists of detection models by hybrid phased modeling and single phased modeling. The integration of the previous two types of modeling methods is to extract the comprehensive fraudulent features in different phases for achieving the capability of early warning in online auctions. To enhance the capability of fraud detection, AntiFraud has implemented a 2-level detection procedure for identifying fraudsters and predicting in which current phase they stay most probably. In addition, the system induces judgment rules of identifying fraudsters from the results of learning processes. In this study, we collected fraudulent transaction histories occurred at Yahoo-Taiwan during 2007-2009 for analyzing the trend of fraud scheme evolution. According to the changes of fraudulent behavior, we construct new behavior models against the new types of frauds as early as possible. To validate the effectiveness of AntiFraud, we also downloaded real data from Yahoo- Taiwan for testing. Our experimental results present the practicality of the system including both the accuracy of fraud detection and the capability of early warning.