Novelty detection, an important topic in machine learning, is the ability of a classification system to differentiate between known and unknown objects in a pattern. Novelty detection has many real-world applications, like in signal processing, pattern recognition, and in the detection of diseases or the prediction of potential faults. For this purpose, different novelty detection models have been proposed.

In this paper, the authors discuss novelty detection in the context of data streams. They consider different aspects of the problem, such as the offline and online phases, the number of classes at each phase (one or more), the number of classifiers used (single or ensemble), the learning task performed (supervised or unsupervised), and many others. Based on these aspects, they propose a well-structured taxonomy of the existing methods, and discuss several applications suitable to novelty detection in data streams, like intrusion detection and text mining. Research challenges and prominent directions for future research are also discussed.

The taxonomy proposed in this paper can be very useful when designing new classification algorithms and machine learning systems. As a result, this paper is expected to be very useful to students who wish to conduct research in novelty detection, as well as to researchers who wish to get a good understanding of the state of the art in this field. Overall, this is a creative and useful work.