Transaction Fraud Detection Based on Total Order Relation and Behaviour Diversity

Abstract

With the popularization of on-line searching, dealings fraud is growing seriously. Therefore, the study on
fraud detection is fascinating and important. a very important approach of police investigation fraud is to extract the
behavior profiles (BPs) of users supported their historical dealings records, then to verify if AN incoming dealings may
be a fraud or not seeable of their BPs. Markoff chain models area unit in style to represent bits per second of users, that
is effective for those users whose dealings behaviors area unit stable comparatively. However, with the event and
popularization of on-line searching, it's additional convenient for users to consume via the net, that diversifies the
transaction behaviors of users. Therefore, Markoff chain models are unsuitable for the illustration of those behaviors. In
this paper, we have a tendency to propose logical graph of BP (LGBP) that may be a total order-based model to
represent the relation of attributes of dealings records. supported LGBP and users’ dealings records, we will work out a
path-based transition chance from AN attribute to a different one. At constant time, we define an data entropy-based
diversity constant so as to characterize the range of dealings behaviors of a user. In addition, we have a tendency to
outline a state transition chance matrix to capture temporal options of transactions of a user. Consequently, we can
construct a BP for every user then use it to verify if an incoming dealings may be a fraud or not. Our experiments over a
real knowledge set illustrate that our technique is healthier than 3 state-of-the-art ones.