Get Started
Research questionHow can unsupervised tabular anomaly detectors provide faithful feature-level explanations without costly post-hoc queries?Unsupervised anomaly detectors must identify unusual records in contaminated, unlabeled samples, but their scores often do not reveal which features caused a point to be flagged. Post-hoc explainers can approximate those reasons only by repeatedly querying the detector.
Machine Learning
Statistical Machine Learning
Latest papersRecent research connected to this question, newest first.Witnesses Explain AnomaliesThe source concerns WAND, an unsupervised detector for tabular anomalies that produces native feature-level explanations during scoring. Evidence comes from comparisons on 47 ADBench datasets against 16 unsupervised baselines, including reported ROC-AUC parity and lower query cost than SHAP, LIME, and ECOD; the evidence is limited to tabular anomaly detection.research paper · Sep 3, 2026
Related questions
How can numerical reconstruction-based anomaly detectors explain which feature ranges cause each alert?How can general-purpose tabular predictors balance adaptation-free accuracy with computational cost?How can graph-level anomaly detection transfer to new domains with little or no target data?How can early-stage chronic kidney disease screening remain accurate and stable with limited labeled data?
Home
Topics
Search
Library