Skip to main content
Thesis defences

PhD Oral Exam - Soheyl Khalilpourazari, Industrial Engineering

Advanced Stochastic Programming and Machine Learning Models for Healthcare Planning, Scheduling, and Prediction Problems

Date & time
Thursday, May 23, 2024
1 p.m. – 4 p.m.

This event is free


School of Graduate Studies


Nadeem Butt

Wheel chair accessible


When studying for a doctoral degree (PhD), candidates submit a thesis that provides a critical review of the current state of knowledge of the thesis subject as well as the student’s own contributions to the subject. The distinguishing criterion of doctoral graduate research is a significant and original contribution to knowledge.

Once accepted, the candidate presents the thesis orally. This oral exam is open to the public.


The increasing demand for global healthcare systems highlights the urgent need for innovative solutions to enhance healthcare delivery. In response to this challenge, this dissertation uses advanced Stochastic Programming (SP) and Machine Learning (ML) methods to introduce significant improvements in the following healthcare areas: appointment scheduling, operating room (OR) planning, modeling and prediction of the COVID-19 pandemic.

In the first paper, we study the healthcare appointment scheduling (AS) problem. Appointment scheduling is a process of allocating a set of time slots to patients with the goal of minimizing the total operational costs. The most significant challenges in appointment scheduling are uncertainties in no-shows, unpunctuality, and service times of patients. Traditional methods often fail to incorporate patient-dependent stochastic service times, patient-and-time-dependent unpunctuality, and patient-and-time-dependent no-shows. This is mainly because the complexity of the resulting appointment scheduling surpasses the capabilities of existing approaches in the literature that highlights the need for more effective modeling and solution approaches. To address this gap, for the first time, we propose a novel stochastic programming model that captures an exponential number of scenarios using a pseudo-polynomial number of variables and constraints without relying on sampling methods. The presented methodology is exact, thus leading to a more precise and efficient optimization of appointment schedules. To enhance the proposed model, we also explore the impact of personalized reminders to minimize the negative impact of no-shows. We show that the generated schedules reduce total costs by 34% on average by incorporating patient-dependent service times, 12% by considering patient-and-time-dependent unpunctuality, and 67% by integrating patient-and-time-dependent no-shows. In addition, we show that personalized reminders have the potential to reduce total costs by 23%.

In the second paper, we study a complicated stochastic operating room (OR) planning problem, a complex and critical healthcare decision-making problem that significantly affects the operational costs of hospitals. The unpredictability of surgical durations poses a considerable challenge to efficient OR planning. Existing models often overlook this source of uncertainty, which leads to either overly optimistic or unnecessarily conservative plans. This paper introduces a novel stochastic programming model that effectively manages the uncertainty in surgical times. This model advances the literature by capturing an exponential number of scenarios in a weekly operating room planning problem without sampling, simplifications, or approximations. The results of the computational experiments revealed that our model obtains feasible solutions with an average optimality gap of 0.78% for instances with 80 surgeries and 1.48E+64 scenarios.

In the third, fourth and fifth papers, we focus on modeling and prediction of the COVID-19 pandemic aiming at developing methodologies that inform and guide public health decisions. In these three papers, we proposed a hybrid reinforcement learning based algorithm as well as two other evolutionary computation based algorithms to forecast the spread of the COVID-19 pandemic. By applying these methods to real-world data from Canada, Quebec, Ontario, France and the U.S., we aim to offer valuable insights into effective pandemic response strategies and emphasize the importance of timely and data-driven decision-making in public health emergencies. These approaches enabled us to predict the pandemic trajectory, which included the number of infected, recovered, and death cases, in the first critical months of the COVID-19 pandemic with high accuracy.

Back to top

© Concordia University