ABSTRACT
This paper introduces a methodological framework for detecting anomalies in non-stationary time series data. It addresses challenges posed by evolving statistical properties and concept drift by using adaptive ellipsoidal decision boundaries, robust covariance estimation, and a lightweight drift detection mechanism.
PAPER · PDF
Loading PDF...
Key findings
Proposes EAD-NS framework for anomaly detection in non-stationary environments.
Combines adaptive ellipsoidal decision boundaries with robust covariance estimation.
Includes a lightweight drift detection mechanism based on Mahalanobis distance statistics.
Limitations & open questions
The paper is a research proposal and thus does not include experimental results.
The effectiveness of the proposed framework is yet to be validated on real-world datasets.