Model-Free Intelligent Diabetes Management Using Machine Learning

• Author / Creator

  Bastani, Meysam

• Each patient with Type-1 diabetes must decide how much insulin to inject before each meal to maintain an acceptable level of blood glucose. The actual injection dose is based on a formula that takes current blood glucose level and the meal size into consideration. While following this insulin regimen, the patient records their insulin injections, blood glucose readings, meal sizes and potentially other information in a diabetes diary. During clinical visits, the diabetologist analyzes these records to decide how best to adjust the patient's insulin formula. This research provides methods from supervised learning and reinforcement learning that automatically adjust this formula using data from a patient's diabetes diary. These methods are evaluated on twenty \emph{in-silico} patients, achieving a performance that is often comparable to that of an expert diabetologist. Our experimental results demonstrate that both supervised learning and reinforcement learning methods appear effective in helping to manage diabetes.

• Subjects / Keywords

  • Actor-critic
  • Supervised learning
  • Reinforcement learning
  • Policy gradient
  • Machine learning
  • Diabetes
  • Type-1 diabetes
  • Insulin dosage adjustment

• Graduation date

  Spring 2014

• Type of Item

  Thesis

• Degree

  Master of Science

• DOI

  https://doi.org/10.7939/R34Q7QX0S

• License

  This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.