I am requesting access to the UK Biobank as part of my graduate research at LIACS, Leiden University, with the objective being the development of a reinforcement learning (RL) based recommender system aimed at mitigating the effects of metabolic syndrome. The system will be designed to deliver personalized recommendations for diet and physical activity, leveraging data-driven insights to improve individual health outcomes.
A key aspect of my thesis is explainability (xAI), which is an essential pillar of trustworthy AI[1]. My objective is to ensure the system provides clear, transparent explanations for its recommendations, thereby fostering trust among users and healthcare professionals[2]. This focus on explainability is carefully balanced with the goal of maintaining a high degree of accuracy in the system’s recommendations.
The primary research question (RQ) will be “How can reinforcement learning algorithms be used to generate personalized recommendations for diet and physical activity for individuals with metabolic syndrome?”. I plan on answering this RQ by applying state-of-the-art methods towards the development of a recommender system based on RL.
The secondary RQ will be “How can xAI be used to increase user comprehension regarding the lifestyle recommendations?”. This question is designed to explore how the integration of xAI techniques can improve user understanding, trust, and transparency to the personalized lifestyle recommendations generated by the system.
Based on my preliminary research, I require access to two data categories: Health-related Outcomes Data and Online Questionnaire Data.
This project is being conducted with the support of Healthbox[3], a collaborative research initiative involving several Dutch universities. The group’s expertise and guidance are instrumental in shaping the methodologies and objectives of this research.
[1] https://arxiv.org/abs/2301.09937
[2] https://arxiv.org/abs/1602.04938
[3] https://www.mijnhealthbox.nl
