The Research Of Click Fraud Detection In Mobile Advertising

With the rapid development of mobile Internet in recent years,the number of mobile users are increasing, the advantages of mobile advertising is accurate, real-time, interactive, diffusion etc. Thus favored by more and more advertisers, and advertisers put ads on mobile device APP or Web site through mobile advertising platform. The most common billing model of advertising is charges per click (CPC), it means user clicks on an ad on website or APP and advertisers pay for the user ad clicks. Mobile advertising platform and publisher (APP developer or website) share the advertising revenue. As the CPC billing model has vulnerabilities, there are click fraud, some advertising publishers in order to reap high profits, will hire many people or using programs to click on their ads on APP or website. This kind of click fraud, caused great harm to mobile Internet advertising industry.This thesis studies the causes of mobile advertising click fraud, and in-depth analysis of domestic and foreign advertising click fraud detection methods and coping strategies. Because click fraud accounts for only a fraction of the total number of clicks, this thesis consider two ways to detect click fraud. The first way is to think of it as an unbalanced classification problem, using SMOTE oversampling technique for dataset equalization, and then using ensemble learning algorithm for detecting click fraud. Experimental results show that the method can detect click fraud effectively. The second way of detecting click fraud is inspired by outlier mining, this thesis presents an improved density-based outlier detection algorithms, and the proposed algorithm is applied to detect click fraud. Experimental results show that the improved algorithm has better results than original density-based outlier mining algorithm.