A New Memory Efficient Technique for Fraud Detection in Web Advertising Networks

The advertising network considered as the middle man in web advertising between advertisers and publishers. This paper presented an intelligent and memory efficient Fraud detection technique with intelligent classification engine to be used by the advertising networks to scan clicks and impressions offline streams happen on publisher side for the purpose of detecting click fraud and impression fraud. The proposed classification technique is based on the proposed data structure for a Scalable Dynamic Counting Bloom Filter (SDCBF). It is a hybrid structure between the Scalable Bloom Filter (SBF) and the Counting Bloom Filter (CBF). It is a variant of the CBF in such a way that, the counter is a dynamic size bit array that can adapt dynamically to its content. Both theoretical analysis and experimental results show that, the investigated technique can achieve minimum space storage with low false positive rate when detecting both duplicate clicks over a sliding window and fast click.