SiD2Re - A novel simulation framework for drifting regression data

Applying supervised machine learning techniques to data streams in non-stationary environments is a challenge that has to be overcome in order to exploit the potentials of machine learning in the context of industrial production. Particularly for regression learning tasks, there is a major need for novel methods to overcome these challenges. In this work we formalize the understanding of data drifts and concept drifts and explain how changes in the process environment can lead to a degradation in model performance. Further, we introduce the Simulator of Drifting Data in Regression problem (SiD2Re) for generating benchmark datasets laying the foundation for comparing new algorithms for drift detection and drift adaptation with respect to various drift characteristics.