Anomaly Types

• Point or Global Anomalies → specific data points
  A point anomaly is where a single datapoint stands out from the expected pattern, range, or norm. In other words, the datapoint is unexpected.

  Example: sklearn.neighbors.LocalOutlierFactor — scikit-learn 1.3.0 documentation

• Collective Anomalies → groups of data (unexpected pattern)
  Where a whole subset of the data deviates from the whole data pattern.

• Contextual Anomalies, where the point deviated based on a selected context

Anomaly Detection libraries

• Tabular data → GitHub - PyOD: Python Outlier Detection
• Time series data → GitHub - TODS: Automated Time-series Outlier Detection System
• Graph data → GitHub - PyGOD: Python Library for Graph Outlier Detection (Anomaly Detection)
• Image data → GitHub - Anomalib: Anomaly detection library comprising SOTA algorithms

Anomaly Detection Resources

• anomaly-detection-resources
• Official Implement of “ADBench: Anomaly Detection Benchmark”
• Paper - ADBench: Anomaly Detection Benchmark
• A practical guide to image-based anomaly detection using Anomalib

ADBench Key Takeaways

• None of the unsupervised methods is statistically better than the others, emphasizing the importance of algorithm selection;
• Semi-supervised methods outperform supervised methods when limited label information is available.

Personal suggestion on selecting an OD algorithm. If you do not know which algorithm to try, go with:

• ECOD: Example of using ECOD for outlier detection
• Isolation Forest: Example of using Isolation Forest for outlier detection scikit-learn