Research
Machine Learning for Risk and Resilience in Complex Systems
Predicting how risk emerges, interacts and cascades, and turning those predictions into decisions that hold up.
- Peer-Reviewed Journal Papers
- Citations
- H-Index
- Papers Reviewed
- Journals Reviewed For
Research Themes
One Core Method. Two Edges. Three Frontiers.
The Core Method
Machine Learning for Risk and Resilience in Complex Systems
Proven in Infrastructure · Project & Asset ManagementPredicting how risk emerges and interacts in systems whose components depend on each other, using multi-output supervised learning built for high-dimensional, non-linear interactions, and forecasting infrastructure project performance with R² of 0.97.
Multi-Output Regression · Deep Neural Networks · Random Forest · XGBoost · Support Vector Machines · Ensemble Methods · Feature Importance
Two Sharpening Edges
Graph-Theoretic Risk Modeling
Networks · Systemic RiskModeling how local disruptions escalate into system-wide failure using network analysis of interdependent risks, pinpointing where intervention does the most to contain a cascade.
Complex Network Analytics (CNA) · Graph Theory · NetworkX · Interdependent Networks · Cascading Failures · Systemic Risk
Optimization-Augmented Machine Learning
Metaheuristics · Decision SupportReplacing exhaustive search with metaheuristics: improving training efficiency by ~90%, carrying predictions through to multi-objective decisions, and releasing the approach as an open-source package with 4,000+ downloads.
Particle Swarm Optimization · Genetic Algorithms · Pattern Search · NSGA-II · Multi-Objective Optimization
Three Applied Frontiers
Hazard Forecasting
ClimateForecasting flood and wildfire hazard from geospatial time series using ConvLSTM and CNN+LSTM architectures, reaching an NSE of 89% on urban flood depth and ~38% better wildfire prediction than baseline models.
ConvLSTM · CNN+LSTM · Geospatial Time Series · Remote Sensing · TFX on Vertex AI
Freight & Air Mobility Risk
Logistics & Supply Chains · Advanced Air MobilityPredicting accident likelihood and impact for drone-based hazmat delivery using a four-model neural-network suite on geospatial data, reaching R² of 0.95 and informing $14M CAD in proposed freight research.
TensorFlow · Neural Networks · Geospatial Features · Hazmat Risk · Supply-Chain Risk
Language Models over Risk Networks
Agentic AI SystemsTesting how reliably language models reason over cascading-failure networks: comparing chain-of-thought, structured and GraphRAG-style prompting on multi-hop dependency graphs, and extending toward small models that call network-analysis tools, running locally.
LLM Evaluation · Multi-Hop Reasoning · GraphRAG · Small Language Models · Agentic Tool Use
Published Research
Peer-Reviewed Publications
Full record on Google Scholar, Web of Science and ORCID.
Journal Articles 8
Real-Time Machine Learning Risk Prediction for Autonomous Drone Operations
A closed-loop ML architecture continuously processes real-time flight, shipment and environmental data to estimate accident likelihood and initiate mitigation responses, with parallel ANN models predicting multi-stakeholder consequence impacts on the public, businesses and customers, closing the loop from prediction to resilience.
ANN · Closed-Loop Architecture · Real-Time Inference · Multi-Output Prediction · Autonomous Decision Support · Consequence Modeling
Moussa, A., E. Hassini, M. Ezzeldin, and W. El-Dakhakhni (2026). “Machine Learning-Enabled Real-Time Risk Prediction and Mitigation in Drone-based Hazmat Delivery.” Research in Transportation Business & Management, 65: 101585. DOI
Heavy Cargo Drones as a Hub-to-Hub Freight Mode
Examines heavy cargo drones as a mode for moving freight between logistics hubs, building on the Smart Freight Centre's work on drones in modern urban freight transportation networks.
Heavy Cargo Drones · Freight Logistics · Advanced Air Mobility · Hub-to-Hub Networks
Moussa, A., E. Hassini, M. Ezzeldin, and W. El-Dakhakhni (2026). “Heavy Cargo Drones for Hub-to-Hub Freight Logistics.” Transportation Planning and Technology. DOI
Graph-Theoretic Machine Learning Pipeline for Cascading Risk Quantification
A three-stage pipeline where complex network theory quantifies team interdependencies as graph features, ML models predict project performance, and network-based and optimization-based mitigation strategies reduce systemic exposure and build resilience into project delivery.
Complex Network Theory · Graph Features · ANN · Random Forest · Cascading Risk · Systemic Risk
Moussa, A., M. Ezzeldin, and W. El-Dakhakhni (2025). “Data-driven Assessment of Complexity-Induced Risks in Infrastructure Projects.” Journal of Construction Engineering and Management. DOI
GA-Tuned ML and Multi-Objective Optimization for Performance Prediction
Genetic algorithm hyperparameter tuning outperforms grid search across all ML models. Predictions then feed NSGA-II multi-objective optimization to generate optimized schedules and risk registers that improve KPIs and build resilience under real-world constraints.
Genetic Algorithms · Hyperparameter Tuning · NSGA-II · Multi-Objective Optimization · Pareto Optimality · Decision Support
Moussa, A., M. Ezzeldin, and W. El-Dakhakhni (2025). “Machine Learning and Optimization Strategies for Infrastructure Projects Risk Management.” Construction Management and Economics. DOI
Supervised ML for Multi-KPI Prediction Under Interacting Systemic Risks
ML models trained on historical project data to simultaneously predict five interdependent performance indicators under the combined effect of interacting and systemic risks, treating multi-output prediction as a first-class modeling problem rather than an afterthought.
ANN · Random Forest · Decision Trees · Multi-KPI Prediction · Systemic Risk · Risk Interactions
Moussa, A., M. Ezzeldin, and W. El-Dakhakhni (2024). “Predicting and Managing Risk Interactions and Systemic Risks in Infrastructure Projects Using Machine Learning.” Automation in Construction, 168: 105836. DOI
Ensemble Machine Learning for Real-Time Dynamic Risk Prediction
Ensemble algorithms trained on injury precursors to generate proactive, real-time safety risk predictions across work zones, moving risk management from reactive lagging indicators to predictive leading indicators.
Ensemble ML · SVM · Random Forest · ANN · Hyperparameter Tuning · Leading Indicators
Gondia, A., Moussa, A., M. Ezzeldin, and W. El-Dakhakhni (2023). “Machine Learning-based Construction Site Dynamic Risk Models.” Technological Forecasting and Social Change, 189: 122347. DOI
Intelligent Features in Construction Robotics: A Review
A structured review of the intelligent capabilities built into construction robots, mapping where sensing, autonomy and decision-making have advanced and where gaps remain.
Robotics · Intelligent Systems · Automation · Literature Review
Attalla, A., O. Attalla, Moussa, A., D. Shafique, S. B. Raean, and T. Hegazy (2023). “Construction Robotics: Review of Intelligent Features.” International Journal of Intelligent Robotics and Applications, 7 (3): 535–555. DOI
Complex Network Theory and Metaheuristic Optimization for Systemic Failure Modeling
The origin of the network-theory edge: contractor interdependence quantified via weighted degree centrality on dynamic and static project networks, correlated with KPI degradation and mitigated through metaheuristic schedule reorganization. The foundation that later fuses with machine learning in the work above.
Complex Network Theory · Weighted Degree Centrality · Dynamic Networks · Metaheuristic Optimization · Systemic Risk
Moussa, A., and W. El-Dakhakhni (2022). “Managing Interdependence-Induced Systemic Risks in Infrastructure Projects.” Journal of Management in Engineering, 38 (5): 04022048. DOI
Conference Papers 2
Heavy Cargo Drones for Last-Mile Delivery
An investigation of heavy cargo drones for last-mile delivery, presented at the Canadian Transportation Research Forum's annual conference in Kelowna.
Heavy Cargo Drones · Last-Mile Delivery · Freight
Moussa, A., E. Hassini, M. Ezzeldin, and W. El-Dakhakhni (2024). “Investigation of Heavy Cargo Drones for Last-Mile Delivery.” Canadian Transportation Research Forum (CTRF), BC, Canada, May 2024.
Prefeasibility of 2,000 MW Remote Hydro Generation
A prefeasibility study for integrating 2,000 MW of hydro generation in remote Northern Ontario, presented at CIGRE Canada.
Energy Infrastructure · Hydro Generation · Prefeasibility
Alrhieh, S., Moussa, A., M. Maghraby, W. El-Dakhakhni, and C. Tang (2023). “Prefeasibility Study for Integrating 2000 MW of Hydro Generation in Remote Northern Ontario.” CIGRE Canada Conference & Exhibition, BC, Canada, September 2023.
Research Philosophy
How I Think About Research
Rigor Without Compromise
Research that cannot be reproduced is not research; it is speculation. Every model, pipeline and result is held to the standard of reproducibility, statistical validity and honest evaluation against strong baselines. Rigor is not a phase of the process; it is the process.
Methods That Travel
A method that only works on the dataset it was built for has limited value. The goal is frameworks that generalize: where the core approach transfers across domains, scales to new data, and survives real-world conditions that were not anticipated at design time.
From Prediction to Resilience
Predictive accuracy is a means, not an end. The full value of a model is realized only when its outputs inform decisions and build resilience, which requires interpretability, uncertainty quantification and integration with the constraints of operational environments.
Reliability as a Design Constraint
Trustworthy systems are not built by accident. Reliability, resilience to distributional shift and graceful failure under unexpected inputs must be designed in from the start, not patched in after deployment reveals the gaps.
Where the Research Is Going
The next frontier is systems that don't just predict accurately, but predict reliably, explain themselves, and remain trustworthy as the world they model keeps changing. From graph-based risk intelligence to language models that reason over risk networks, the through-line is the same: AI that holds up under pressure, at scale, in production.
Let's Build Something That Matters
If you have a decision that depends on predicting risk in a complex system, I'd like to hear about it.