Appier shared real-world research analysis results from its network, demonstrating the AI-based model’s ability to prevent ad fraud.

The company conducted research using more than 4 billion campaign data points, including ad clicks and app install information, over a four-month period from May to August 2017. The study found that models based on artificial intelligence were able to analyze fraudulent install patterns twice as fast as existing rule-based models.

The advantage of AI-based models is that they can detect fraudulent install patterns that are difficult for traditional models to detect. For example, one fraud pattern that Appier discovered is called “chameleon,” which refers to a publisher who initially disguises themselves as a legitimate ad publisher but later generates fraudulent installs. Another fraud pattern is called “inventory burst,” which involves counting abnormally high inventory when there is not a proper level of in-app activity on the install advertiser’s application.

“Fraud has become a major threat to the online advertising industry, and is expected to cost advertisers billions of dollars in the coming years,” said Joe Su, Appier’s chief technology officer. “Traditional rules-based methods for detecting fraud and mitigating its impact have limitations. Appier believes that AI-based models are far more effective, and in fact, we can see the benefits of the AI approach being evident after just four months of analyzing the network,” he added.

Existing rule-based models typically only look at one to three dimensions and operate on already known negative install patterns according to rules programmed by humans. On the other hand, AI-based models not only examine data in over 80 dimensions, but also provide self-learning capabilities, allowing them to detect new suspicious patterns that were not seen before.

The CTO added, “Just like cyber fraud or financial fraud, fraudulent installs are becoming more sophisticated and constantly evolving, so it is important to quickly identify new threats and minimize their impact. Traditional rule-based approaches cannot keep up with the fraudsters’ modus operandi, and AI-based models are needed to effectively track evolving fraudulent install patterns.”