An outlier is an observation in a dataset that appears to be inconsistent with the remainder of that set of data. Anomaly detection can be defined as a process that will detect such outliers. Anomaly detection can be categorized into the following types based on the percentage of labelled data:

• Supervised anomaly detection is characterized by the following:
  • Labels available for both normal data and anomalies
  • Similar to rare class mining/imbalanced classification

• Unsupervised anomaly detection (outlier detection):
  • No labels; training set = normal + abnormal data,
  • Assumption: anomalies are very rare
• Semi-supervised anomaly detection (novelty detection):
  • Only normal data available to train
  • The algorithm learns on normal data only