COMPARATIVE STUDY BETWEEN STATISTICAL FRAUD DETECTION METHODS ON ECOMMERCE NETWORKS

In the eCommerce industry, the problem of consumer fraud is becoming an increasingly troublesome problem posing a multitude of challenges. Large eCommerce firms, such as Alibaba, eBay, Amazon (to name a few) serve the needs of hundreds of millions customers. On a daily basis, a large number of transactions are made and recorded, as more consumers are becoming susceptible to fraudulent behaviour provided the scale of operations. This research aims to develop a robust real-time fraud detection algorithm based on the construction of a behavioural reference model. Time series features pertaining to a data set of customers on a large eCommerce service are collected and assigned to two classes (non-fraudulent or fraudulent) and modelled using a general stochastic model. The classification performance of this model is compared with more conventional classification techniques, such as logistic regression, tree-based methods, and Na誰ve Bayesian Classifier.M.A.S