Active Learning for One-Class Classification Using Two One-Class Classifiers

  • 2019-01-10 12:36:25
  • Patrick Schlachter, Bin Yang
  • 2

Abstract

This paper introduces a novel, generic active learning method for one-classclassification. Active learning methods play an important role to reduce theefforts of manual labeling in the field of machine learning. Although manyactive learning approaches have been proposed during the last years, most ofthem are restricted on binary or multi-class problems. One-class classifiersuse samples from only one class, the so-called target class, during trainingand hence require special active learning strategies. The few strategiesproposed for one-class classification either suffer from their limitation onspecific one-class classifiers or their performance depends on particularassumptions about datasets like imbalance. Our proposed method bases on usingtwo one-class classifiers, one for the desired target class and one for theso-called outlier class. It allows to invent new query strategies, to usebinary query strategies and to define simple stopping criteria. Based on thenew method, two query strategies are proposed. The provided experiments comparethe proposed approach with known strategies on various datasets and showimproved results in almost all situations.

 

Introduction (beta)

None

 

Conclusion (beta)

None