> ML_LITERATURE // LIU-TING-ZHOU-2008-ISOLATION-FOREST_v1.0
Isolation Forest
Fei Tony Liu, Kai Ming Ting, Zhi-Hua Zhou · IEEE International Conference on Data Mining (ICDM) (2008)
foundational2008industry-standardthirdPartyReproduced
Principal Contribution
Inverted the anomaly detection paradigm by isolating anomalies directly via random tree path length rather than profiling normal points.
Operational Relevance
The global standard baseline for tabular anomaly detection, financial fraud monitoring, and industrial sensor fault screening.
Assumptions
- Anomalies are few and have attribute values distinct from normal instances, requiring fewer random splits to isolate
Limitations
- Axis-aligned cuts produce artifacts in empty regions between diagonal clusters (addressed by Extended Isolation Forest)