Dijital, anlık, uyarlanabilir müdahalelerin dizayn ve kişisilleştirilmesine yönelik bir sistem.

Adverse and suboptimal health behaviors and chronic diseases are responsible from a substantial majority of deaths globally. Studies show that personalized support programs yield better results in overcoming these undesired behaviors and diseases. Digital, just-in-time, adaptive interventions are mobile phone-based notifications that are being used to support people wherever and whenever needed in coping with the health problem. In this study, a framework is proposed for design and personalization of such interventions. The design part targets intervention designers and allows them to configure interventions that address specific needs of a particular health problem or population. The personalization part presents a reinforcement learning based mechanism to optimize intervention delivery strategies with respect to timing, frequency and type of interventions. Specifically, two reinforcement learning models, namely intervention-selection and opportune-moment-identification, are employed simultaneously. The models are fed with data obtained pertaining to people's long-term and momentary contexts. While the intervention-selection model adapts the intervention delivery with respect to type and frequency, the opportune-moment-identification model tries to find the most opportune moment to send interventions. Two improvements over the standard reinforcement learning algorithms are proposed to boost the learning performance. First, a customized version of eligibility traces is employed to reward past actions throughout the agent's trajectory in a selective manner. Second, the transfer learning method is utilized to reuse knowledge across multiple learning environments. The design and personalization modules of the proposed approach are validation individually. For the design part, it is shown that the proposed approach addresses the requirements of the intervention design specifications extracted from the extant literature. It is also shown that the design mechanism was utilized to design interventions for a real-life case program. The personalization part is validated mainly via a simulated case-study. Four personas are simulated with differentiating parameters in their daily activities, preferences on specific intervention types and attitude towards the targeted behavior. The results show that the improved algorithms yield better results compared to the standard versions and capture the simulation variations associated to the personas. A small-scale real-life case study has also been conducted utilizing a preliminary version of the proposed personalization method. Better results were obtained by the proposed approach compared to the base algorithm and a fixed intervention delivery schedule.Ph.D. - Doctoral Progra