Click Stream Data Analysis for Online Fraud Detection in E-Commerce

Web services became the integration part of our life at the present time including advertisement on various web pages. Many e-commerce companies generate advertisement revenue by selling clicks (it is known as Pay-Per-Click model). In this model, e-commerce company is paid for each time an advertisement link on its website is clicked leading to the sponsoring company’s content. However, some of these companies inflate the number of clicks their sites generate. Generation of such invalid clicks either by humans or software with the intension to get fraudulently money is known as click fraud. In this article we show how the click fraud can be unmasked using various time features (e.g., period of the day and the day of the week when a user’s (that is identified by his IP address) clicking occur). We combine several different time features into a timeprint. We use machine learning methods in a number of experiments to get an understanding of to what extent time prints can be used for identifying click fraud. The obtained results show that timeprints indeed can be a useful tool for the improvement of the quality of click fraud analysis.