Record-Driven Blockwise Anomaly Detection in Weakly Dependent Time Series
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
We consider a real-valued time series {Xt}t≥1 and aim at detecting changepoints understood as persistent departures from a reference local regime. We propose a distribution-free method that combines (i) a local power statistic {Pt} derived from local asymptotic normality (LAN) testing framework, (ii) blockwise aggregation of exceedance events, and (iii) a principled calibration of the tolerance parameter d using record theory (under the independent and identically distributed (i.i.d.) , linear-drift (LDM), and YangNevzorov models). Under β-mixing, block exceedance counts admit a binomial approximation, yielding tractable decision rules that separate normal behavior, transient anomalies, and genuine changepoints. Simulations illustrate the practical performance and robustness of the approach.
Fields of Study & Subjects
Advanced Statistical Process Monitoring
Anomaly Detection Techniques and Applications
Fault Detection and Control Systems
Publication details
- Journal / source
- HAL
- Publication date
- 19 March 2026
- Language
- EN
- Document type
- Preprint