EAS Doctoral Dissertation Defense by Fatemeh Salboukh
LIB 426
Dr. Lance Fiondella
lfiondella@umassd.edu
Abstract:
Modern systems support critical services in transportation, energy, communication, finance, defense, healthcare, and other domains where failures or prolonged disruptions can lead to economic loss, degraded service, safety risks, and loss of public trust. As these systems become more complex, interconnected, and data-driven, ensuring that they remain reliable during normal operation and resilient under disruptive conditions become increasingly challenging. Previous research on system reliability has developed statistical and machine learning models for defect prediction and defect-prone module classification. Similarly, previous research on system resilience has developed metrics and models to assess how systems absorb, recover from, and adapt after disruptions. These approaches provide important foundations for system assessment, but many remain limited by insufficient treatment of causal and conditional dependencies among system metrics or insufficient flexibility for nonlinear, dynamic, and time-varying conditions. As a result, existing models may not fully support prediction and decision-making when defects, covariates, disruptions, and recovery processes evolve over time. To address these limitations, this dissertation develops statistical and machine learning models for system reliability and resilience prediction. For system reliability, this dissertation develops a causal inference-inspired feature selection approach for defect-prone module classification, allowing cause-effect relationships and conditional dependencies among system metrics to guide the selection of informative predictors. For system resilience, this dissertation develops predictive models for forecasting system performance under dynamic conditions, including models that account for delayed effects, nonlinear relationships, and time-varying covariates.
The proposed models are validated using benchmark and real-world datasets related to defect-prone module classification and resilience prediction. Goodness-of-fit and classification performance measures are used to compare the proposed approaches with established statistical and machine learning baselines. The results show that incorporating causal and conditional dependencies, delayed effects, nonlinear relationships, and covariate behavior improves predictive performance across the problems considered. These findings demonstrate that the proposed models can support practical decisions such as identifying high-risk modules, planning recovery actions, and forecasting long-term system performance. Overall, this dissertation provides flexible and interpretable modeling approaches that advance system reliability and resilience assessment for complex systems operating in uncertain and changing environments.
ADVISOR(S): Dr. Lance Fiondella, Department of Electrical & Computer Engineering (lfiondella@umassd.edu)
COMMITTEE MEMBERS:
- Dr. Alfa Heryudono, Department of Mathematics
- Dr. Gokhan Kul, Department of Computer & Information Science
- Dr. Ruolin Zhou, Department of Electrical & Computer Engineering
- Dr. Hong Liu, Department of Electrical & Computer Engineering
NOTE: All EAS Students are ENCOURAGED to attend.