Changing A/B-testing into A&B-testing with exceptional model mining

A/B-testing is a well-known method, often used in marketing, to compare two variants of the same product and select the most successful variant. This variant is retained, while the other variant gets discarded. But with all the data gathered with A/B-testing, isn’t it possible to acquire more information about the two products, besides determining the variant that is the most successful on average?

That question was the starting point of a cooperation between Assistant Professor Wouter Duivesteijn, and Full Professor Mykola Pechenizkiy of TU Eindhoven’s Department of Mathematics and Computer Science, StudyPortals, and a team of four Bachelor students. A cooperation that eventually even led to an accepted paper for the Applied Data Science Track at ECML PKDD 2017, one of the largest European conferences on machine learning and data mining.