A Proposed Future Activity Prediction Model for Services Supporting Sustainable Behavior without Strain

Abstract

This paper proposes a future behavior prediction model aimed at realizing services that support desirable behavior changes and continuations, such as health promotion and learning, for the users. We assume that the time slot in which the users can comfortably implement new behaviors is when the probability of scheduled or routine behaviors such as sleep and meals occurring is low. However, the time slots for these scheduled or routine behaviors change daily, and their duration also varies due to factors such as fatigue, making it impossible to predict them using traditional methods based on pre-planned schedules and statistical predictions. Therefore, we propose an Encoder-Decoder model that predicts the time slot in which the user can comfortably implement new behaviors, taking into account three elements: schedule, routine behavior, and fatigue. We evaluated the proposed model using two datasets and confirmed that it can predict with higher accuracy than statistical methods and can adapt to daily changes.