One-class classification

• tries to identify objects of a specific class among all objects
  • using only a training set composed of that one class
• note that a normal Classification problem has access to all classes in the training set.
• applications in outlier-detection, anomaly detection and novelty detection
• several one class algorithms :
  • one-class Guassian
    • assume data came from a guassian distribution -> maximize likelihood
  • one-class k-means
    • have thresholds to decide if new feature vector similar to existing samples
  • one-class kNN
    • similar process to that of k-means
  • one-class SVM
    • either tries to seperate all training examples from the origin and maximize distance from the hyperplane to the origin
    • or tries to obtain a spherical boundary around the data by minimizing the volume of this hypersphere