Skip to main content

Mixed Effects Models for Predicting Anomaly-Detector Error Rates

Researcher: Roy Maxion

Research Area: Security of Cyber-Physical Systems

Abstract

Scope: We will use collected data to derive a model for the magnitude of the effects of multiple diverse factors on anomaly-detector accuracy, thereby determining which factors have the most influence, either contributing to or detracting from detector accuracy. The model will be validated using additional data, not collected for estimating the model. The validated model will serve as a trustworthy predictor of detection accuracy, showing which factors most affect detector performance.

Outcomes: A validated model for predicting detector error rates. A method for constructing such a model for detectors in general domains.