In the talk, we address the types of fraud in RTB (Real Time Bidding) ecosystem (bots, ad stacking, spoof sites). Then we discuss what kind of fraud can be resolved by means of various approaches including machine learning, e.g. modified bid clustering for good traffic (human) and bad (bot). We also discuss which clustering method is better, which way of learning (supervised/unsupervised) is suitable, how feature selection may help in terms of fighting fraud. As for the technical part, we discuss the impact of different parameters (e.g., size of learning sample, number of Google Cloud Engine machines needed) and possible ways of computational optimisation.
展开▼
