Outlier Detection with Outlier Classification

Outlier Classification

Supervised Learning

• Samples were labeled by domain experts.
• Model normal objects and report those not matching the model as Outliers
• Model Outliers and treat those not matching the model as normal

Challenges are:

• Imbalanced classes
  • boost outlier class by Oversampling and maybe Undersampling of normal objects
• catch as many outliers as possible
  • Recall is more important than Accuracy

Unsupervised Learning

• No labels available
• Use Clustering method to cluster data
  • samples not falling in any cluster are outliers

Challenges are:

• Samples outside of clusters might not be outliers
• Costly
• Hard to distinguish from noise
• cant detect collective outliers as they will most likely be treated as just another cluster

Semi-Supervised Learning

• only few labels available
• if some normal objects are labeled
  • use these and the proximate ulabeled object to train a model for normal objects
  • report non fitting objects as outliers
• if some outlier objects are labeled
  • may not cover all possible types of outliers
  • thus use models from Unsupervised Learning to help recognize normal objects