Integrating Novel Class Detection with Classification for Concept-Drifting Data Streams

In a typical data stream classification task, it is assumed that the total number of classes are fixed. This assumption may not be valid in a real streaming environment, where new classes may evolve.