NASA’s new dark energy space telescope can also detect killer asteroids
technologyreview.com·3h ago
AI Research
arXiv CS.AI·Jun 12·5 min read
CRAFTIIF: Cross-Resolution Analytic Four-Type Interpretable Isolation Forest for Multivariate Time Series Anomaly Detection
✦ Why It Matters
Engineers can leverage CRAFTIIF for robust anomaly detection across diverse multivariate time series applications.
Key Takeaways
- Utilize CRAFTIIF for comprehensive anomaly detection in time series data.
- Leverage four distinct Isolation Forests for targeted anomaly types.
- Apply adaptive Otsu/MAD thresholds for automatic calibration.
- Evaluate performance using the mTSBench benchmark datasets.
How It Works
CRAFTIIF employs a unique approach by generating 500 random analytic wavelet features from four families: Morlet, DOG, Haar, and Coiflet. Each feature set is tailored to detect a specific type of anomaly.
These features are then processed through five structured Isolation Forests, which are specialized for each anomaly type, plus a meta-Isolation Forest that handles compound anomalies. This design allows for direct attribution of detected anomalies to their respective types based on the branches that fire during the detection process.