This project explores the use of reinforcement learning (RL) techniques to determine optimal insulin dosing for patients with Type I diabetes. Utilizing a Type I diabetes simulator, we approached the problem as a Markov Decision Process (MDP) and applied two RL strategies: Action Value Function Approximation (AVFA) and Deep Q-learning (DQN).
The link to our write up can be found here
Diabetes is a chronic condition characterized by high blood sugar levels due to the body's inability to produce sufficient insulin. This project aims to automate insulin dosing decisions using RL to maintain blood glucose levels within a safe range.
We adapted the diabetes simulator to represent our problem as an MDP and explored it using AVFA and DQN strategies. Our methods showed significant improvements in controlling glucose levels compared to a randomized baseline agent.
Both AVFA and DQN models outperformed the random-action agent, maintaining blood glucose levels closer to optimal levels, with room for improvement during mealtime disturbances.
Our code and further documentation are available in our GitHub repository.
The project was a collaborative effort between Yichen Jiang and Emmy Thamakaison, focusing on literature review, algorithm implementation, simulation, result analysis, and reporting.
For detailed information, installation instructions, and usage examples, please refer to our GitHub repository.