Employing Environmental Data and Machine Learning to Improve Mobile Health Receptivity

Abstract

Behavioral intervention strategies can be enhanced by recognizing human activities using eHealth technologies. As we find after a thorough literature review, activity spotting and added insights may be used to detect daily routines inferring receptivity for mobile notifications similar to just-in-time support. Towards this end, this work develops a model, using machine learning, to analyze the motivation of digital mental health users that answer self-assessment questions in their everyday lives through an intelligent mobile application. A uniform and extensible sequence prediction model combining environmental data with everyday activities has been created and validated for proof of concept through an experiment. We find that the reported receptivity is not sequentially predictable on its own, the mean error and standard deviation are only slightly below by-chance comparison. Nevertheless, predicting the upcoming activity shows to cover about 39% of the day (up to 58% in the best case) and can be linked to user individual intervention preferences to indirectly find an opportune moment of receptivity. Therefore, we introduce an application comprising the influences of sensor data on activities and intervention thresholds, as well as allowing for preferred events on a weekly basis. As a result of combining those multiple approaches, promising avenues for innovative behavioral assessments are possible. Identifying and segmenting the appropriate set of activities is key. Consequently, deliberate and thoughtful design lays the foundation for further development within research projects by extending the activity weighting process or introducing a model reinforcement.