Did We Actually Fix It? An Independent Adversarial Stress-Test of Post-Point-Adjustment Evaluation Metrics for Time-Series Anomaly Detection
Zongye Lyu
stat.ML, cs.LG
Submitted: 2026-07-12
Comments: 16 pages, 4 figures
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- A Problem-Oriented Taxonomy of Evaluation Metrics for Time Series Anomaly Detection
- Towards Unbiased Evaluation of Time-series Anomaly Detector
- Multivariate Time Series Anomaly Detection: Fancy Algorithms and Flawed Evaluation Methodology
- Precision and Recall for Time Series
- Formally Exploring Time-Series Anomaly Detection Evaluation Metrics
- SimAD: A Simple Dissimilarity-based Approach for Time Series Anomaly Detection
Related papers
- Behavior of prediction performance metrics with rare events
- Optimal Estimation of Generic Dynamics by Path-Dependent Neural Jump ODEs
- A Posterior-Dynamics Framework for Imaging Inverse Problems with Pretrained Diffusion Priors
- Scalable Posterior Uncertainty for Flexible Density-Based Clustering
- A Transport Geometry of Benign Overfitting
- Efficient quadratic entropy with distance sketches