ABSTRACT
Novelty detection is a learning task that consists of the identification of new or unknown concepts that the learning system is not aware of during training. This is one of the fundamental requirements of a good classification or identification system, since sometimes the test data contain information about concepts that were not known at the time of training the model. In timedependent applications, novelty detection represents an important challenge, since concepts are hardly ever constant. This chapter surveys the major approaches for the detection of novelty in data streams.1
