• A
  • A
  • A
  • ABC
  • ABC
  • ABC
  • А
  • А
  • А
  • А
  • А
Regular version of the site

Working paper

DEDPUL: Difference-of-Estimated-Densities-based Positive-Unlabeled Learning

Positive-Unlabeled Learning is an analog of supervised binary classification for the case when the Negative (N) sample in the training set is contaminated with latent instances of the Positive (P) class and hence is Unlabeled (U). We develop DEDPUL1 , a method able to solve two problems: first, to estimate the proportions of the mixing components (P and N) in U, and then, to classify U. The main steps are to reduce dimensionality of the data using Non-Traditional Classifier [11], to estimate densities of the reduced data in both P and U samples, and to apply the Bayes rule to density ratio. We experimentally show that DEDPUL outperforms current state-of-the-art methods for both problems. Additionally, we improve current state-of-the-art method for PU Classification [16].