The term dataset shift originates from the fields of artificial intelligence, big data, smart data and digital transformation. Dataset shift describes a change in the data that artificial intelligence or algorithms work with, for example. This means that the data used to develop and train models suddenly differs from the data that the system receives later in use.
A simple example from practice: A company develops a system for recognising fraud transactions in online retail. The system is trained using data from recent years. But suddenly customer behaviour changes, for example because many people switch to a certain payment method due to a new trend. However, the previous data does not describe this new behaviour. As a result, the model often makes the wrong decisions.
Dataset shift can lead to AI models or data analyses becoming less reliable or even delivering completely incorrect results. Companies must therefore regularly check whether their models are still working correctly with new, changed data. This is the only way to utilise the full potential of artificial intelligence and big data and avoid costly wrong decisions.