Dual-Tier Feature Hierarchy in Behavioral Drift Detection: Universal Indicators vs Domain Specialists
Abstract
Feature importance analysis reveals a dual-tier hierarchy: Energy Distance (f1536) as universal first responder (~5x importance in cross-domain transfer), and Hodge/sheaf decomposition (f196) as domain specialist (Cohen's d 5-120 in LLM-native training). Constructs a feature importance heatmap across 29 models revealing module activation patterns.
Fields of Study & Subjects
Data Stream Mining Techniques
Anomaly Detection Techniques and Applications
Fault Detection and Control Systems
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 1 March 2026
- Language
- EN
- Document type
- Preprint