Towards Minimally Domain-Dependent and Privacy-Preserving Architecture and Algorithms for Digital Me Services: EdNet and MIMIC-III Experiments

“Digital Me” refers to a service that mirrors an individual's goals, monitors and predicts their status, and provides recommendations to improve the status. This study investigates the architecture and algorithms designed to predict user status and indicate actions, relying solely on the user's data. The objective is to reduce dependency on domain-specific models while promoting the sustainable and privacy-preserving accumulation of data. To validate the proposed architecture and algorithms, we employed the EdNet dataset in the education sector and the MIMIC-III dataset in healthcare. We developed algorithms that recommend activities to optimize users' goal achievement. These algorithms follow a fundamental principle applicable to general Digital Me services: recommending actions most likely to improve the user's subsequent state based on their probability of success and expected outcome. Additionally, we demonstrate the viability of creating effective health prediction algorithms through personal federated learning. This enhances privacy by avoiding centralizing sensitive health data and storing it on individual devices or private clouds.