This study aims to determine the best practices and provide a model of the technical solutions that can effectively and systematically limit fraudulent transactions of online orders in e-commerce services, using the methods of analytical mining and case studies. Based on a process of fraud prevention and detection performed in the e-business Dangdang, Inc., a leading online retailer in China, twelve identifying features of fraudulent order data were extracted and compiled into a feature matrix. Logistic regression with this matrix was then used to build a model to judge if an order was fraudulent. The model was tested using various order data with machine learning techniques to meet the requirements of being effective, correct, adaptive, and persistent. Then an online detection and prevention schema was established and the hypothesis of so-called Behavior Pattern Change Assumption (BPCA) was proven. The results show the model can detect 94% of fraudulent orders. The Anti-fraud Schema System established for Dangdang is shown to be the best model for the determination and prevention of fraudulent behaviors in the e-commerce services.