Active Learning for One-Class Classification
This thesis explores the intersection of two areas of research in the field of machine learning. First, it involves active learning, a family of methods that aim to either reduce the cost of manually labeling data or to optimize the labeling process towards a more useful data set, in the optic of machine learning purposes. Second, this thesis is targeted towards the problem of one-class classification, where data from only one class is available, and the goal of classification if to categorize future data that either does or doesn't belong to this class. The main contribution of the thesis is a method where active learning is applied in the context of one-class classification.