BuzzCity has been working with the Singapore Management University and the National University of Singapore to improve its ability to identify and manage click fraud.

In a series of crowd-sourcing and development events, researchers analysed data including publisher ID, campaign ID, handset, IP address and the time of the click to try and establish how fraudsters resemble humans and evade detection.
Click fraudsters, the researchers found, have reached a high level of sophistication to help them mimic legitimate users in large volumes. Fraud detection is challenging for many data mining and machine learning algorithms, they said, because the practise involves many variables and may only account for a small fraction of all the clicks.

The new predictive models established by the researchers outline several combinations of criteria that should enable mobile marketers to identify abnormal behaviour and reduce false alarms. An ‘ensemble model’ that uses many different modelling techniques can lead to better predictions and performance than individual algorithm, they found.
“Click fraud effectively enables fraudsters to make a small sum of money from each fraudulent click, which can add up to quite a large sum of money if done enough times,” said Dr KF Lai, CEO of BuzzCity. “The results of this study are critical to maintaining the integrity of the performance of mobile advertising campaigns for the industry as a whole.”
The findings have been published in the Journal of Machine Learning Research and are currently being incorporated into BuzzCity’s mobile advertising network.