Projects & Portfolio

🚑Healthcare Systems & Pandemic Response

I develop data-driven models and optimization approaches to strengthen healthcare decision-making, particularly during pandemics and in pharmaceutical supply chains. My healthcare research includes:

• Optimal Patient Allocation During Infectious Disease Outbreaks

I proposed an integrated forecasting–optimization framework to allocate patients across hospitals with limited capacity during infectious disease outbreaks. The approach combines a hierarchical time series model for forecasting hospital bed demand with a Susceptible–Infected–Recovered (SIR) model to capture metapopulation dynamics. The resulting problem was formulated as a mixed-integer nonlinear program and validated using real Covid-19 data from all 67 counties in Florida.

The framework reduced unmet hospital bed demand by 30% statewide, demonstrating the value of optimized patient transfers in mitigating regional surges. Analysis revealed that the hierarchical model performed best in high-capacity regions, while univariate models were more effective in low-capacity regions, leading to different county-level demand distributions despite similar statewide outcomes. Sensitivity analysis showed that the maximum transfer distance was not a major performance driver, and the framework recommended 20-day to monthly planning intervals for effective decision-making.

This study illustrates how data-driven forecasting and optimization can improve healthcare system resilience, reduce strain during pandemics, and guide policymakers in tailoring strategies to regional capacity constraints.

• Optimal Covid-19 Vaccine Distribution and Waste Management

I developed a sustainable fuzzy multi-objective optimization model to jointly manage Covid-19 vaccine distribution and the medical waste generated by vaccination programs. The model integrates a capacitated, multi-depot vehicle routing problem (VRP) with heterogeneous vehicles and introduces a fuzzy time-window mechanism to capture vaccination centers’ satisfaction levels and service priorities. The optimization was solved using augmented ε-constraint, TH, and LP-Metric methods and applied to a real-world medical supply chain in Iran.

The results demonstrated that adjusting vaccination centers’ service times increased satisfaction levels up to 0.991, while emissions were reduced by 17.1%. Sensitivity analysis further showed that vehicle hiring cost variations could shift total network costs by up to ±18.7%, underscoring the model’s ability to evaluate trade-offs between cost, environmental impact, and service quality. This study provides practical guidance for sustainable vaccination logistics, helping decision-makers align health outcomes with environmental goals.

• Sustainable Pharmaceutical Supply Chain Design

I designed a multi-objective optimization model for a closed-loop pharmaceutical supply chain that integrates recycling and sustainability criteria. The approach applied the Analytical Hierarchy Process (AHP) to rank green manufacturers and was solved using the LP-metric method in GAMS.

In a real-world case study in Iran, the model established 10 manufacturers, tracked multi-period inventory and recycling flows of up to 172 units per period, and reduced overall costs by 11% while simultaneously improving environmental and social outcomes. This work underscores the importance of sustainable practices in healthcare logistics and pharmaceutical management, providing decision-makers with practical strategies to balance cost efficiency with sustainability.

• Blood Supply Chain Optimization 

I developed a novel mixed-integer nonlinear programming (MINLP) model for blood supply chains that integrates location, allocation, inventory management, and finite-capacity queueing systems. The model simultaneously minimizes costs and maximizes donor satisfaction by reducing waiting times at collection centers.

In a real-world case study, the system managed a 62% shortage in the first period (312 unmet units of 500 capacity), then successfully collected 5,900 units in the second period (about 6.9% above demand) and carried over 200 units (3.4%) for storage in the third period. By capturing shortages, surpluses, and recycling opportunities, the model demonstrated how optimized blood supply chain networks can improve efficiency, resilience, and donor experience.

• Predicting Covid-19 Survival with Blood Characteristics

I conducted a machine learning study to predict survival outcomes of 306 Covid-19 patients from Tangji Hospital in Wuhan using clinical blood indicators. The dataset was imbalanced, so I applied oversampling methods to address class imbalance before training the models. I compared multiple algorithms, including Logistic Regression, Decision Trees, KNN, SVM, Random Forest, and AdaBoost.

The results showed that ensemble and tree-based models achieved prediction accuracies exceeding 90%, with Random Forest, SVM, and AdaBoost performing best. Critical predictors included age, lactate dehydrogenase (LD), and C-reactive protein (CRP). This work highlights how non-medical, data-driven approaches can complement healthcare systems by providing early insights into patient outcomes during pandemics.

🚘Transportation Safety Research

I investigate data-driven methods to improve roadway safety and crash prediction, with a focus on context-specific risk factors, statistical modeling, and causal analysis. My transportation-related research includes:

• Detecting Roadway Features with GIS

I am conducting an ongoing project that applies ArcGIS Pro to detect roadway safety features  (specifically guardrails, medians, and shoulders) across Florida roadways. The analysis distinguishes between on-system and off-system roads, with case studies in Leon, Alachua, Liberty, Walton, and Broward counties.

Preliminary results show that the method achieves about 70% detection precision in identifying roadway features. This work supports infrastructure safety assessments by enhancing the accuracy of roadway feature inventories, helping transportation agencies more effectively evaluate risks and prioritize safety improvements.

• Causal Effects of Special Events on Rural Crash Risks

I examined how special events influence the likelihood of pass-by crashes in rural Florida using regularized logistic regression with group lasso feature selection. The study integrated over 25,000 crash records with roadway attributes and socioeconomic factors across two FDOT districts during 2021–2022.

Results revealed that special events increased pass-by crash risks by up to 60% in certain counties, with substantial county-level differences in how events shaped travel behavior and safety outcomes. These findings provide transportation planners with actionable insights into where and when event-driven travel poses heightened crash risks, supporting more targeted interventions for rural safety management.

• Diagnostic Testing for Highway Safety Manual Models

I developed a diagnostic test for evaluating negative binomial regression assumptions in Safety Performance Functions (SPFs) from the Highway Safety Manual. Using three years of crash data from FDOT District 4, I introduced a spatial sampling approach that ensures regression models meet key statistical assumptions such as linearity, overdispersion, and zero inflation.

The analysis identified 20-mile subregions as optimal for modeling crashes of all injury levels, while 60-mile subregions provided better validity for severe injury crashes. This approach reduced overdispersion by 48%, improved linearity by 16%, and lowered overall model error by 2%, making FDOT’s crash predictions more statistically reliable.

• Context-Specific Thresholds for Crash Predictions

I developed a functional statistical model to establish minimum traffic volume (AADT) thresholds for Safety Performance Functions (SPFs) across roadway types, segment lengths, context classifications, and crash severities. This context-aware method reduced SPF prediction errors by up to 89%, ensuring more reliable crash predictions on Florida roadways and helping the Florida Department of Transportation (FDOT) avoid misclassifying high-risk sites.

In a complementary project with FDOT’s safety department, I applied statistical and exploratory data analysis to refine existing crash prediction models, achieving up to a 45% increase in prediction accuracy. Together, these projects provide FDOT with more precise tools to identify dangerous locations, allocate safety resources, and prevent severe crashes across the state.

🌪️Disaster Management: 

I focus on developing advanced models and decision-support tools to improve disaster preparedness and response. My key projects include:

•  Compound Hazards: Pandemics and Geophysical Disasters

I developed a spatiotemporal mixed-effects model to study how hurricanes amplify pandemic risks, using Covid-19 and Hurricane Sally in Florida as a case study. The model analyzed infection trajectories across 67 counties, accounting for demographic and social vulnerability factors.

Results revealed that daily case counts doubled (100% increase) in some counties immediately after landfall, with average increases of up to 40% in vulnerable regions. Key predictors of elevated transmission included group quarters (+3.9% daily cases per 1% increase), disability prevalence (+4.9%), and multi-unit housing (+1.4–1.8%).

This work highlights how community-level vulnerabilities shape disease trajectories during compound disasters, offering actionable insights for public health planning and disaster preparedness.

• Humanitarian Supply Chain Optimization

I formulated a multi-objective MINLP model to optimize disaster relief operations by determining the location of relief centers, assigning transportation methods, and allocating both perishable and non-perishable relief items. The model was designed to minimize travel distance, total costs, and delivery time while preventing resource waste. Small test cases were solved in GAMS, and for large-scale applications I implemented the Grasshopper Optimization Algorithm (GOA).

In the 2019 Iran flood case study, the model achieved significant improvements in efficiency. Relief items traveled a total of 5.2 million units of distance with a minimized cost of 12.4 billion Rials (≈$295,000). The average delivery time from relief centers to demand points was reduced to just 1.2 hours, while the allocation strategy eliminated waste, with zero perished items recorded. The solution coordinated a mixed fleet of 205 vehicles for imperishable goods and 90 vehicles for perishables, ensuring the timely delivery of over 5 million standardized relief items across the affected regions. 

This work illustrates how advanced optimization can enhance disaster response by cutting costs, reducing delays, and ensuring vital resources are delivered efficiently and without waste.

• Role of NGOs and Disruptions in Relief Operations

I extended humanitarian supply chain models to incorporate medical treatment, NGO participation, and disruptions in relief centers, solved with a two-stage fuzzy-stochastic approach and the Grasshopper Optimization Algorithm (GOA). The model was applied to a Tehran case study to evaluate disaster response under uncertainty.

Results showed that NGO cooperation reduced total costs by 13.6% (from 5.88M to 5.08M) and significantly lowered the number of relief items that had to be procured by the government (from 61 down to 0 as NGOs increased contributions). By contrast, relief center disruptions raised costs by nearly 20%, highlighting the importance of resilient infrastructure.

This study demonstrates how multi-actor collaboration and resilience planning can improve efficiency and equity in humanitarian relief operations.

🌎Sustainable Transportation & Energy

I also conduct research on sustainable transportation systems and energy infrastructure, with a focus on supporting the transition to cleaner mobility.

• Electric Vehicle Charging Station Location

I developed a bi-objective mixed-integer linear programming model to determine the optimal locations and numbers of electric vehicle (EV) charging stations under budget and disruption constraints. The model incorporates multiple charger types, accounts for station breakdowns, and was solved using a Lagrangian relaxation method.

In a real-world Tehran case study, the model recommended installing 15 chargers across 2 optimal sites out of 11 possible candidates. This solution used 96% of the available budget while ensuring that 80% of EV users were fully served, with only 20% needing to travel beyond their tolerable distance. Sensitivity analyses further revealed how reallocating chargers versus opening new stations could reduce costs without sacrificing user satisfaction.

This research provides practical strategies for policymakers to expand EV infrastructure, balance costs with user accessibility, and strengthen resilience against disruptions.