Date of Award
4-2025
Degree Name
Doctor of Philosophy
Department
Computer Science
First Advisor
Shameek Bhattacharjee, Ph.D.
Second Advisor
Ajay Gupta, Ph.D.
Third Advisor
Alvis Fong, Ph.D.
Fourth Advisor
Kevin Lee, Ph.D.
Abstract
Cyber-Physical Systems (CPS) rely on anomaly-based detection methods to ensure the integrity and security of critical infrastructures such as smart grids, smart water metering systems, and advanced metering infrastructures (AMI). Anomaly detection methods are commonly used to identify deviations from normal system behavior by establishing learned profiles and thresholdbased distinctions between benign and anomalous events. However, conventional frameworks often fail to account for adversarial data poisoning attacks, unlabeled unsafe events, and environmental noise—factors that distort training data, degrade detection accuracy, and increase false alarms. This dissertation proposes a resilient learning framework that mitigates these biases by integrating quantile regression, M-estimation theory, and majorization theory into a unified approach. Specifically, we introduce quantile-modified M-estimators and redescending M-estimators with robust empirical risk functions, effectively balancing the trade-off between false alarms and missed detections—an issue exacerbated by imbalanced event probabilities in training data. While the likelihood of a data poisoning attack is low, our framework is designed to be resilient under both adversarial and benign conditions. It not only maintains robustness against potential attacks but also optimally determines anomaly detection thresholds in the absence of adversarial manipulation, ensuring reliable performance even when the system operates under normal conditions. Moreover, we provide theoretical derivations to identify the most suitable loss function from the family of M-estimators, ensuring optimal resilience against data biases. Empirical evaluations on real-world smart CPS datasets, including smart energy and water meters, validate the proposed framework’s effectiveness in mitigating training data attacks and improving anomaly detection performance. By advancing resilient machine learning techniques for anomaly detection, this dissertation contributes to the broader field of trustworthy and robust AI for CPS security.
Access Setting
Dissertation-Open Access
Recommended Citation
Abedzadeh, Sahar, "Resilient learning for Anomaly Detection in Smart living CPS" (2025). Dissertations. 4145.
https://scholarworks.wmich.edu/dissertations/4145