ABSTRACT
In this chapter, we start with the de’nitions of novel class and existing classes. en we state the assumptions based on which the novel class detection algorithm works. We illustrate the concept of novel class with an example, and introduce several terms such as used space and unused spaces. We then discuss the three major parts in novel class detection process: (1) saving the inventory of used spaces during training, (2) outlier detection and ’ltering, and (3) computing cohesion among outliers and separating the outliers from the training data. We also show how this technique can be made e¨cient by raw data reduction using clustering.
