A Walk Through the Isolation Forest
summary
Isolation Forests for anomaly detection are explained, covering the original algorithm and the Entropy Isolation Forest extension, with attention to their intuition, parallelism and performance. The mathematical formulation is analyzed and extended to provide native support for categorical features and missing values, and the talk includes a GitHub implementation to illustrate these extensions.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 0:43Anomaly detection basics
- 2:06Why hard
- 3:14Supervised vs unsupervised
- 4:39Isolation Forest basics
- 6:24How trees work
- 7:27Anomaly scoring
- 8:57Splitting and isolation
- 10:43Training and sampling
- 11:48Visualizing anomalies
- 13:14Implementation and tools
- 14:17Categorical data extension
- 16:47Frequency handling
- 18:31Missing values impact
- 20:41Bias in missing data
- 22:46Benchmarking and use cases
- 23:29Q&A
related talks
33mUnsupervised Anomaly Detection with Isolation Forest
26mAnomaly Detection System for the CERN Cloud
01h 16mAnomaly Detection
39mSniffing Out the Strange — A Look at Anomaly Detection
22mAnomaly Detection: The Bad Boy of Data Science
29mAnomaly Detection
45mAnomaly detection with TensorFlow
34mML-Bound Anomaly Detection: Experiences Beyond Commodity Observability