Concept Drift Detection and Adaptation in IoT Data Stream Analytics

Dynamic nature of the IoT data can often cause machine learning model degradation, which can lead to analysis and actions failures. In order to address this challenge, it is of utmost importance to detect drifts in the data that can cause unwanted behavior of the system, and to update models accordingly. This paper gives an overview of different methods for addressing concept drift detection and adaptation, demonstration of some of the methods using open source libraries, and a design for a component for smart adaptive system for streaming data analysis that uses presented techniques to ensure high model performance.