Abstract: Transaction fraud is predominant nowadays as many people prefer shopping through online. This made the research on detecting fraud to become an important part of research at present times. Based on records of transaction of several users, Bp (Behavioral Profile) is extracted and then verification takes place in order to find the frauds occurring. Usage of Markov models does not give relevant results, thus making unsuitable to use in such scenarios. Attributes from several transaction records are classified in the proposed system to represent a graph like structure. The probability of path from one attribute to another is calculated along with computation of diversity of several different users behaviors in transaction. For every user a Bp is constructed and is verified whether the transaction is allowed incoming or not to detect the fraud. The experimental results show the efficacy of the system.