A Density-based Approach for Positive and Unlabeled Learning

抄録

Positive and Unlabeled learning (PU learning) is a machine learning approach that focuses on generating a two-class classification model using only a set of positive examples, and a set of unlabeled examples. Various techniques have been proposed for PU learning. Most of the techniques try to detect a group of reliables negative examples from the given unlabeled examples. Then, a classification model can be incrementally built. In this paper, we propose a new technique for detecting the reliable negative examples based on the density of examples in the search space.