Anomaly Detection

Detecting Anomalies

Anomaly detection identifies unusual patterns.

Statistical Methods

Z-score: |z| > 3 flags outliers. IQR method: outside Q1-1.5IQR to Q3+1.5IQR.

Distribution fitting: data far from fitted distribution.

Isolation Forest

Tree-based: isolates anomalies quickly. sklearn.ensemble.IsolationForest.

contamination parameter sets expected anomaly rate.

One-Class SVM

Learns normal data boundary. Novelty detection mode for new data.

Key Takeaways

1. Statistical methods are simple baseline
2. Isolation Forest handles high dimensions
3. One-class SVM learns normal boundary